# AI & the Future of Knowledge Work MOC ## Synthesis additions (2026-10-17 — employment and AI: a vendor's report on its own workforce) - [[3. Resources/Readwise/Articles/How Employment Is Changing at Anthropic]] — Anthropic's research post on its own staff: junior share of technical departments reported falling from 21% to 14%, about 80% of interactive agent-session time spent waiting for a person, meta-work rising from 10% to 22% of sessions. First-party vendor self-report: the post's headline finding is stronger than its own caveat that adopters were already on an upward trend, and several figures carry no method. - [[3. Resources/Permanent Notes/Comparing Adopters With Non-Adopters Credits the Tool for the Adopters' Prior Trend]] — a vendor's own sentence says adopters of its first fully autonomous agent were already on a steep upward trajectory, so an adopter comparison overstates the tool's effect. ## Synthesis additions (2026-10-15 — agent speed claims and personal software) - [[3. Resources/Permanent Notes/A Ticket-to-Merge Clock Stops Before the Fix Reaches Users]] — ask where an agent-speed claim's clock starts and stops (Brad Vincent thread) - [[3. Resources/Permanent Notes/A Nightly Agent Rebase Makes a Personal Fork Cheap Enough to Keep]] — maintenance, not writing, kills personal software; a nightly rebase removes that cost for small patch sets (Crawshaw) - [[3. Resources/Readwise/Tweets/My Agents Took a WordPress Support Ticket to a Merged Fix in Just Over 2 Hours]] — Brad Vincent's thread: a WordPress support ticket to a merged fix in just over 2 hours with agents; self-reported by the plugin's maker, with the release a week later. ## Synthesis additions (2026-10-04 — building agent skills as software) - [[3. Resources/Literature Notes/Almost Timely News - Highly Opinionated Method of Building AI Agent Skills (2026-10-04)]] — Penn's four skill directives: orchestration-only root file, as little AI as possible, one job per skill, tests before release. - [[3. Resources/Permanent Notes/Deterministic Skill Steps Belong in Scripts Not the Model]] — a step with one right answer runs as a script, leaving the model the language work. - [[3. Resources/Permanent Notes/Skill Activation Needs a Labeled Should-Fire Test Set]] — a skill description is a classifier, so it ships only after passing labeled should-fire and should-not-fire prompts. ## Synthesis additions (2026-10-01 — engineering careers under AI: apprenticeship, isolation, and review) - [[3. Resources/Readwise/Articles/AI Makes Critical Thinking Harder to Build]] — LeadDev opinion piece (no byline, no cited data): AI coding tools remove friction that taught junior engineers to weigh tradeoffs, so leaders should build critical-thinking practice into code review, RFCs, and career frameworks. - [[3. Resources/Readwise/Articles/Your Error Budgets Don’t Know AI Exists]] — LeadDev opinion piece (no byline): error budgets measure consumed failure, not whether owners understand a change; AI cuts the cost of producing changes but not of understanding them. - [[3. Resources/Readwise/Articles/The Senior Engineer Death Spiral]] — Sunil Pai describes a spiral of over-ambitious project, silence, secret overwork, and burnout, and says remote work and coding agents make it easier to enter and harder for others to spot. ## Synthesis additions (2026-10-01 — AI in knowledge-work practice) - [[3. Resources/Readwise/Articles/The AI Proposal That Lost Us a Client]] — Duncan Isaaksen-Loxton (The Agency Hour; both speakers sell related services): AI cut research on eight client sites from about two days to 30 minutes, yet a client rejected the 13-page proposal as sounding AI-made. - [[3. Resources/Readwise/Articles/Almost Timely News 🗞️ How I Use AI as a Keynote Speaker & Facilitator]] — Christopher Penn's newsletter (self-reported, also marketing for Trust Insights) describes using AI before, during, and after keynotes and workshops, arguing AI removes the old cost of packing a session with material. - [[3. Resources/Readwise/Articles/This AI Does Your ENTIRE Systematic Literature Review in Minutes]] — Andy Stapleton demos the AnswerThis review tool and calls it among the best tested, while saying every step needs checking; no accuracy test is shown. - [[3. Resources/Readwise/Articles/Why Claude Can’t Be Your PM (Yet) Anthropic CPO Panel]] — Lenny's Podcast panel: Anthropic's Ami Vora and Mike Krieger, questioned by Dan Shipper, argue faster building raises the value of the PM role; vendor-affiliated speakers, anecdotal evidence. ## Synthesis additions (2026-10-01 — AI-built software and durable agent memory) - [[3. Resources/Readwise/Articles/Lessons Learned From Vibe Coding Plugins]] — WP Minute meetup: Patrick Gallagher of GridPane (a site sponsor) argues AI tools now produce working plugins but hallucinate, and proposes multi-model red-teaming plus an accountable human; benchmark figures are unverified. - [[3. Resources/Readwise/Articles/Stop Losing Your Best Ideas in AI Chats - GIVE CHAT MEMORY]] — Zsolt Viczian shows a branch, commit, and pull-request pattern where ChatGPT reads and writes a private GitHub repository so brainstorms persist; the video also promotes his paid StoryOS cohort. ## Synthesis additions (2026-09-23 — prompts as governed documentation assets) - [[3. Resources/Permanent Notes/Documentation Prompts Are Documentation Assets Governed Like Code]] — Whitney Burkes (Write the Docs Berlin): shared doc prompts live in a repo, pass review, and are tested across three or four codebases before they count as done. ## Synthesis additions (2026-09-23 — open weights as the precondition for local fallback) - [[3. Resources/Literature Notes/Facebook and Meta's Llama Language Model 2]] — the 2023 origin point of the sovereignty argument: a commercially licensed open-weight model is what makes local inference a real fallback, with 301 derivatives inside 24 hours as the adoption signal. ## Synthesis additions (2026-09-09 — two AI manifestos: vendor and practitioner) - [[The AI Realist Manifesto - Beyond the Vibe]] — Pantheon's internal memo: a four-quadrant map (AI Realist = optimist + pragmatist), three principles, a W3Techs "None" reversal as the vibecoding data point, and a role table for how designer, content, developer, and agent jobs change. - [[The AI-Augmented Software Development Manifesto - Ronan Berder]] — Agile Manifesto rewritten for cheap code generation: human agency and taste over agentic autopilot; "LOC count, agent swarms, and token burn are not progress." ## Synthesis additions (2026-09-05 — where the human sits in an AI pipeline) - [[3. Resources/Permanent Notes/Where a Human Stands in an AI Pipeline Decides Judgment or Volume]] — Same tool, same scaffold, different outcome depending on whether the human sits downstream of generation or at the review gate. ## Synthesis additions (2026-09-05 — AI displacement, credible sources) - [[3. Resources/Permanent Notes/The Most Reliable Evidence a Skill Has Been Automated Comes From Whoever Taught It]] — A teacher conceding their own taught skill is now automated is stronger evidence than an outside analyst's claim, because the concession costs the teacher something real. ## Synthesis additions (2026-09-05 — AI vendor supply-chain concentration risk) - [[3. Resources/Permanent Notes/A Dependency Can Be Severed by a Transaction Between Two Other Parties]] — OpenAI's Cursor model-supply cutoff after the SpaceX acquisition, as the natural experiment for model-swap risk in AI coding tools. ## Synthesis additions (2026-09-05 — disclosure as the fix for unevaluable process claims) - [[3. Resources/Permanent Notes/A Quality Label Without a Quantity Is Not a Claim You Can Check]] — a WP newsletter's own build-notes disclosure (model, tokens, review time) shows how a cheap magnitude turns an unevaluable quality label into a checkable one ## Synthesis additions (2026-09-05 — prompt/instruction maintenance discipline) - [[3. Resources/Permanent Notes/Prompt Rules Decay With Model Versions Date Every Never]] — a prompt rule correcting one model generation's bias can amplify the opposite bias in the next; date every behavior-correcting rule ## Synthesis additions (2026-09-05 — style-rule design for AI writing) - [[3. Resources/Permanent Notes/Ban the Move Not the Word for Style Rules That Outlast Banned-Word Lists]] — banning a substitution (metaphor for available literal phrase) outlasts banning specific vocabulary ## Synthesis additions (2026-09-05 — review layer for AI-authored edits) - [[3. Resources/Permanent Notes/AI Edits Should Land as Attributable Tracked Changes Not Direct Mutation]] — Unattended AI document edits should land as attributable tracked changes a person accepts per-change, not as direct mutation — the maker/checker separation applied to prose. ## Synthesis additions (2026-09-05 — AI mediation changes the value of owned content) - [[3. Resources/Permanent Notes/Owning a Website Is Becoming an Accuracy Control Rather Than a Traffic Channel]] — an owned site may function as an accuracy correction for AI answers even as its click-through traffic declines ## Synthesis additions (2026-09-05 — legible AI-agent review output) - [[3. Resources/Permanent Notes/Severity Labels Make Review Rank Legible When No Prefix Means Required]] — a no-prefix-means-required severity scheme is the cheap fix that makes an agent's ranked review output legible to a human who didn't watch it run ## Synthesis additions (2026-09-05 — misreading fear when an input becomes cheap) - [[3. Resources/Permanent Notes/When an Input Becomes Cheap Doubling Down on It Is the Fear Response]] — doubling down on effort/throughput is the reflexive fear response exactly when AI has made that input cheap ## Synthesis additions (2026-09-05 — openness as extensibility rather than licensing) - [[3. Resources/Permanent Notes/Openness Is a Surface Property Not a License Property]] — Whether a system is genuinely open turns on whether third parties can extend it without permission, not on what license governs the code or weights — Apache vs. Android as the contrast case. ## Synthesis additions (2026-09-05 — launch marketing shaping developer usage) - [[3. Resources/Permanent Notes/Launch Examples Function as Usage Instructions]] — A vendor's launch-blog examples function as de facto usage instructions regardless of the tool's actual comparative strengths (OpenAI 3D-demo case). ## Synthesis additions (2026-08-30 — integrity under automation) - [[3. Resources/Permanent Notes/Moral Image Is Contingent on the Crowd and Moral Identity Is Not]] — Grant's distinction: a constraint built on being seen as good dissolves exactly when defection becomes common, which is when it was needed. - [[3. Resources/Permanent Notes/An Anti-Misconduct Measure That Advertises Prevalence Increases the Misconduct]] — An honesty pledge before an exam doubled cheating — applies to compliance training and security-awareness campaigns alike. ## Synthesis additions (2026-08-24 — backlog gap-fill (second pass)) - [[3. Resources/Permanent Notes/State a Model Cannot Hold Belongs in Version-Controlled Files]] — the context window is not storage, and a diff is the only honest record of what an agent removed. - [[3. Resources/Permanent Notes/Search Returns Opinion About a Change Not the Change Itself]] — ranking favours commentary over primary records, so an agent reasons from takes rather than sources. - [[3. Resources/Permanent Notes/A Metric Can Drop Because the World Changed or Because Your Instrument Broke]] — the check that has to precede every externally-collected finding. - [[3. Resources/Permanent Notes/The Demand for an Oracle Precedes Any Technology Capable of Being One]] — over-trust in confident answers is often downstream of a prior epistemology rather than induced by the tool. - [[3. Resources/Permanent Notes/The Model Is a Thin Swappable Layer Over a Thick Infrastructure Codebase]] — the architecture that makes a model swap a days-long task rather than a quarter-long one. - [[3. Resources/Permanent Notes/Routine Work Provided In-Workday Recovery That Automation Removes]] — the cost of automating the low-effort work people were quietly recovering during. ## Synthesis additions (2026-08-24 — commit late, verify always) - [[3. Resources/Permanent Notes/Abstract Only After Two Real Implementations Exist]] — the counterweight to reversibility advice: keeping options open has an up-front cost, and two implementations is the first moment an abstraction can be checked rather than guessed. - [[3. Resources/Permanent Notes/Automation Removes the Social Checks a Process Silently Depended On]] — the review-side companion: generation got cheap, and the informal check that used to precede review disappeared with it. - [[3. Resources/Permanent Notes/Model Evidence Is Bound to the Product Surface and Settings]] — model evaluations need the surface and runtime configuration that produced them. - [[3. Resources/Permanent Notes/When Review Cannot Scale Correctness Must Move Into the Toolchain]] — the keystone of this thread: once output exceeds what anyone can inspect, quality becomes a property of what the tooling refuses, not of how carefully people look. ## Synthesis additions (2026-08-12 — process disclosure) - [[3. Resources/Permanent Notes/AI Involvement Labels Disclose Process Not Quality]] — treats AI authorship marks as production-process disclosure rather than a quality or truth score. ## Synthesis additions (2026-08-11 — hidden formation time) - [[3. Resources/Permanent Notes/Expertise Conceals the Duration That Produced It]] — restores the intermediate learning steps hidden by expert performance. ## Synthesis additions (2026-08-11 — selection proxies) - [[3. Resources/Permanent Notes/Credential Screens Select for the Proxy Not the Work]] — separates success at a credential gate from demonstrated capability for the actual job. ## Synthesis additions (2026-08-11 — team capability) - [[3. Resources/Permanent Notes/Individual Agent Adoption Does Not Become Team Capability by Aggregation]] — requires shared review and reusable practice before individual tool use counts as organizational capability. ## Synthesis additions (2026-08-11 — task and job effects) - [[3. Resources/Permanent Notes/Task Automation Does Not Predict Job Elimination]] — separates a capability benchmark's task-level result from the demand and task-bundle dynamics that determine employment. ## Synthesis additions (2026-08-11 — disposable code boundary) - [[3. Resources/Permanent Notes/Disposable Code Should Produce Evidence Not Become the Product]] — permits cheap generated probes while preserving review for every artifact that crosses into production. Router-format index (rewritten 2026-08-01; prose version archived in `.trash/`). One line per note: open it if the line answers your question. Labor and formation consequences of AI — the operational-controls side lives in [[AI Agent Safety MOC]]. ## Core Question When AI compresses the entry tier of knowledge work, where does the cost of producing senior judgment go — and who decides whether the remaining humans are augmented or coerced? ## Pipeline compression & apprenticeship - [[3. Resources/Permanent Notes/AI Compresses the Knowledge-Worker Pipeline From Both Ends]] — the foundational data: $200+/user/mo spend up, ages-22-25 headcount −13% in AI-exposed roles - [[3. Resources/Permanent Notes/AI Tools Don't Eliminate the Apprenticeship Layer They Relocate Its Cost]] — entry tasks go, but the formation they funded still has to be paid for - [[3. Resources/Permanent Notes/Code Review Is a Human Knowledge-Transfer Ritual Not Just a Quality Gate]] — agent-to-agent PRs hollow out the loop that makes seniors (McCue) - [[3. Resources/Permanent Notes/Stability Is the Precondition for Formation Across Domains]] — formation needs the stable time compression removes - [[3. Resources/Permanent Notes/AI Should Raise Student Challenge, Not Lower Expectations]] — scaffolding = augmentation; task replacement = lost practice - [[3. Resources/Literature Notes/You Ask, I Answer — Enterprise AI And The Junior Talent Pipeline]] — Penn: retrain juniors for AI oversight, not execution - [[3. Resources/Permanent Notes/DevRel Programs Take Three to Five Years to Show Results]] — the timeline AI both shortens and threatens - [[3. Resources/Permanent Notes/Maintainer Attention Is the Scarce Resource in Open Source]] — the OSS instance of pipeline scarcity - [[3. Resources/Permanent Notes/Forced Exploration Builds a Model an Instant Answer Cannot]] — the transferable model comes from the search, so exploration now has to be chosen rather than endured. ## Power & deployment configuration - [[3. Resources/Permanent Notes/Productivity Gains Become Workload Inflation Without a Relief Policy]] — explains how faster work becomes an unsustainable output baseline when organizations capture all saved capacity. - [[3. Resources/Permanent Notes/The Reverse Centaur Pattern - Humans Forced to Work Like Machines]] — same tool augments or coerces depending on who sets the pace - [[3. Resources/Permanent Notes/Default-On AI Activation Inflates Adoption Metrics at the Cost of Consent]] — the vendor mechanism pushing reverse-centaur defaults - [[3. Resources/Permanent Notes/AI's Failure Mode Is Careless Adoption Not the Technology]] — "AI employee" labels lower reviewer scrutiny (Wiles/NYT) - [[3. Resources/Permanent Notes/Heavy AI Use Removes the Fatigue Signal That Tells You to Stop]] — burnout arrives invisibly when the mental-fatigue cue is muted - [[3. Resources/Literature Notes/Gutenberg Changelog 133 (Gutenberg 23.6 Release and WordPress 7.1)]] — Imokol's first-person account of AI adoption compelled by competitive shipping-speed pressure rather than preference - [[3. Resources/Permanent Notes/Competitive Velocity Converts Tool Adoption Into a Condition]] — names the mechanism: when output speed is visibly comparable across competitors, non-adoption itself becomes the costly choice - [[3. Resources/Literature Notes/AI Productivity Is Burning Out Your Best Engineers]] — backfilled Lit Note; its optimizing-one-metric-hides-invisible-work claim was already promoted from this exact source as the existing Prevented Work permanent. - [[3. Resources/Literature Notes/Governance Is the Platform Problem Worth Solving]] — LeadDev/Harness AI-agent-governance panel; its agent-identity-delegation claim links to the existing evergreen identity-is-the-control-plane permanent, which already cites this source. - [[3. Resources/Permanent Notes/Governance Funding Follows the Decision-Makers Existing KPIs]] — a governance proposal gets funded when its benefit is translated into a metric the specific decision-maker already tracks, not when the case is made for governance in the abstract. ## Judgment, review, and taste as the new bottleneck - [[3. Resources/Permanent Notes/Human Review Capacity Sets the Agent WIP Limit]] — caps parallel execution at the point completed runs begin queueing for human judgment. - [[3. Resources/Permanent Notes/AI Activity Metrics Measure Tool Throughput Not Engineering Value]] — artifact counts track agent generation after value moves into outcomes and judgment. - [[3. Resources/Permanent Notes/Managers Should Code Off the Critical Path]] — keeps AI-enabled manager coding additive by preventing managerial commits from gating delivery. - [[3. Resources/Permanent Notes/The Bottleneck Moved From Code Generation to Review and Judgment]] — generation cheap → problem-translation and review now bind - [[3. Resources/Permanent Notes/Implementation Is Cheap So Taste Becomes the Bottleneck]] — the selection-side companion: knowing which cheap output is good - [[3. Resources/Permanent Notes/When Answers Are Cheap Understanding Becomes the Scarce Good]] — the human's job moves up a level: judge the filter's output - [[3. Resources/Permanent Notes/Cheap Inspection Collapses the Signal-Substance Gap]] — when checking is free, verifiable substance beats polished signal - [[3. Resources/Permanent Notes/Sampling a Fraction of a Source Produces False Authority Over the Whole]] — Forte's "Gap": connectors read 0.5–5% but answer as if whole - [[3. Resources/Permanent Notes/AI Search Inherits the Trust and Fragility of the Sites It Cites]] — a 13-word planted snippet flips outputs across a query cluster - [[3. Resources/Literature Notes/Almost Timely News - Can AI Be An Investment Expert (2026-08-02)]] — an evidence-based refusal to a question the user visibly wanted answered yes is evidence the instrument isn't sycophantic; connects to the existing permanent note on this source rather than duplicating it - [[3. Resources/Literature Notes/The Complexity of Measuring Value in Knowledge Work]] — why visible-output metrics miss judgment value - [[3. Resources/Literature Notes/The Shift from Clarity to Ambiguity in Knowledge Work]] — value resists output proxies even before AI obscures it - [[3. Resources/Literature Notes/The Problem with Streaming and the Case for Physical Media]] — revocable platform access vs. owned media; cross-linked with the Almost Timely companion piece below (same publish window, same argument from the opposite side). - [[3. Resources/Permanent Notes/Don't Outsource the Thing That Makes It Yours]] — ownership means keeping a copy or a capability you control, because access you don't own can be revoked by whoever holds the switch. - [[3. Resources/Literature Notes/Almost Timely News - The Biggest Problem with AI Today]] — paired with the streaming/physical-media piece as a parallel ownership-vs-outsourced-judgment argument. - [[3. Resources/Permanent Notes/Outsourcing Judgment Removes the Accountable Decision-Maker]] — delegating executive function to AI rather than leading it leaves no human owning the decision, so accountability quietly erodes. - [[3. Resources/Literature Notes/Your SDLC Is Your Context Engineering (LeadDev)]] — agentic-SDLC redesign case study; its spec-is-the-bottleneck claim links to the existing agent-specification permanent rather than duplicating it. - [[3. Resources/Permanent Notes/Self-Review Fails at the Task Level Not Just the PR Level]] — a single agent session can now span an entire task before a PR opens, so per-PR reviewer separation alone no longer prevents self-review. - [[3. Resources/Literature Notes/Effective Verification for Humans and Agents]] — LeadDev/CodeRabbit panel; two of five claims already covered by the existing oracle-first-verification and delegated-accountability permanents. - [[3. Resources/Permanent Notes/Agentic Coding Speed Gains Are Unevenly Available Across Codebases]] — greenfield codebases capture agentic speed gains faster than legacy, regulated ones, because the fastest verification patterns need a freedom to redesign that regulated environments don't have. - [[3. Resources/Permanent Notes/A Change Summary Can Catch Structural Problems a Diff Review Would Miss]] — a natural-language "what changed and why" summary can catch an agent's architectural overreach that a line-by-line diff review would likely pass over. - [[3. Resources/Literature Notes/Strategy Before Software]] — WP Product Talk panel; its build-cost-collapse claim already sits in the existing taste-is-the-bottleneck permanent, which already cites this episode. - [[3. Resources/Permanent Notes/A Working Strategy Is Measured by What It Lets a Team Refuse]] — a strategy proves itself by how reliably it lets a team say no to a technically-easy feature, not by how polished the document reads. - [[3. Resources/Permanent Notes/Issue Structure Raises Agent Fix Quality More Than Model Choice]] — shape the task before escalating the model; the "how to verify" line is what converts an open request into one with a stop condition. - [[3. Resources/Permanent Notes/A Repeatable Workflow Becomes Capacity Only When It Becomes a Skill]] — inputs, guardrails, and a human review step are what turn a prompt into delegable capacity. - [[3. Resources/Permanent Notes/Tool-Mastery Tasks Do Not Substitute for Deep Work]] — AI widens the gap by making the completion-signal-rich half faster. ## Amplifier thesis & the harness - [[3. Resources/Permanent Notes/Volatile Context Belongs at the Event Where It Changes]] — uses lifecycle hooks to refresh facts that go stale inside a long-lived agent session. - [[3. Resources/Permanent Notes/Persistent Memory Turns Health Metrics Into Coaching Context]] — a personal-agent case where durable state changes isolated measurements into longitudinal interpretation. - [[3. Resources/Permanent Notes/AI Amplifies Existing Bottlenecks Rather Than Creating Capability]] — DORA: AI 10x's whatever was already the constraint - [[3. Resources/Permanent Notes/Weak Links Cap Aggregate Gains Even Under Widespread Automation]] — 10–20x code vs 30% releases: the unautomated step caps the gain - [[3. Resources/Permanent Notes/AI Quality Gaps Are Policy Problems Not Capability Problems]] — most "model isn't good enough" is a harness/policy gap (Awais) - [[3. Resources/Permanent Notes/The Harness Now Outpaces the Model as the Source of Agent Capability Gains]] — durable leverage sits in the loop, not the model - [[3. Resources/Permanent Notes/Long-Horizon AI Shifts the Bottleneck From Model Capability to Memory and Effort Scaffolding]] — the plateau constraint is scaffolding - [[3. Resources/Permanent Notes/AI Capability in a Domain Advances Through a Fixed Three-Stage Arc]] — assist → assisted → autonomous; stage forecasts trajectory - [[3. Resources/Permanent Notes/A Standardized Harness Turns Model Access Into a Second-Sourceable Supply Decision]] — swappable endpoints make bans a failover - [[3. Resources/Permanent Notes/Clean Code Is the Human-AI Interface]] — enforceable standards are the control surface for agent collaboration - [[3. Resources/Permanent Notes/Spec-Driven Development Turns Throwaway Prompts Into Durable System Memory]] — versioned specs persist intent that prompts lose - [[3. Resources/Permanent Notes/Encoding What Good Looks Like Is What Makes Agent Verification Trustworthy]] — checked-in spec = the external standard for agent self-review - [[3. Resources/Permanent Notes/Eval Quality Is the Rate-Limiter on Safe Agent Autonomy]] — evals, not the model, cap safe autonomy - [[3. Resources/Permanent Notes/AI's Leverage Comes From the Data Foundation Not the Model]] — fragmented data → confident noise at scale no prompt fixes - [[3. Resources/Permanent Notes/Vague Prompts Are Expensive Specificity Cuts AI Error and Cost Together]] — the request-level cost lever - [[3. Resources/Literature Notes/Lit · Fable 5 prompting — effort and memory]] — effort dial + durable lesson store for long-horizon work - [[3. Resources/Literature Notes/Setting Up a Prompt Engineering Workflow]] — prompting as iterative testable process - [[3. Resources/Literature Notes/I need you to hear me out (it's REALLY good)]] — t3.gg: harness orchestration design can outweigh model choice - [[3. Resources/Permanent Notes/Perm · Values-explicit prompting closes AI-human writing gap]] — explicit values + audience evals narrow output variance - [[3. Resources/Permanent Notes/Zet · Multi-model routing beats single-model for creative work]] — route subtasks by model fit - [[3. Resources/Permanent Notes/Perm · AI agents can disappear overnight if a provider changes policy]] — provider policy changes are dependency failures; test fallbacks - [[3. Resources/Permanent Notes/Perm · Live-edge observability replaces static logs as the monitoring standard]] — real-time detection as operating environment - [[3. Resources/Permanent Notes/Zet · AI providers require output format standardization]] — provider-specific output needs one normalized contract - [[3. Resources/Literature Notes/The epic story of Markdown (The Vergecast)]] — Gruber on Markdown's forgiving error semantics as a distinct, LLM-specific reason for its dominance over stricter formats. - [[3. Resources/Permanent Notes/A Formats Error Tolerance Not Its Expressiveness Predicts LLM Output Preference]] — a format that degrades gracefully on a small error beats a more expressive format that fails outright, as an LLM's default output choice. - [[3. Resources/Permanent Notes/Ask The Codebase Questions Before Editing It]] — Cherny's adoption-ladder claim: interrogating a codebase before editing it builds calibration faster than jumping straight to changes; distinct claim from the same source as the existing feedback-loop permanent above. - [[3. Resources/Literature Notes/How to Use AI Deep Research to Improve WordPress Plugins]] — using AI deep research as a plugin-improvement technique, with two distinct scaffolding claims promoted. - [[3. Resources/Permanent Notes/Benchmark-Gated Adoption Lets You Ship Techniques You Don't Fully Understand]] — a passing benchmark, not full theoretical understanding, is what makes an unfamiliar optimization technique safe to adopt. - [[3. Resources/Permanent Notes/Cap Deep-Research Depth at One or Two Levels]] — bounding each research branch to one or two levels of narrowing keeps exploration converging on something shippable instead of an open-ended rabbit hole. - [[3. Resources/Literature Notes/Don't Use Next.js Without These 7 Skills]] — its context-not-capability claim links to the existing Context Not Constraints permanent; only the agent-facing-docs claim was new. - [[3. Resources/Permanent Notes/Write Agent-Facing Docs Assuming the Agent Wont Fetch the Reference]] — put the critical instruction inline at the point of use rather than trusting an agent to follow a link out to a separate reference file. - [[3. Resources/Literature Notes/Claude Tag Now Reads Even More of the Room]] — Claude Tag's shift to channel-context-aware Slack behavior. - [[3. Resources/Permanent Notes/A Proactive Agent Earns Trust by Staying Quiet Not by Responding Often]] — an ambient assistant earns trust by defaulting to silence and clearing a confidence bar before speaking, not by maximizing how often it responds. - [[3. Resources/Literature Notes/Firecrawl + Convex]] — Firecrawl's Convex component tutorial for search/scrape/durable crawl. - [[3. Resources/Permanent Notes/A Serverless Function That Cannot Wait Must Hand the Job to a Callback]] — when a serverless function can't stay open long enough to finish a job, the job has to be handed off and completed through a callback. - [[3. Resources/Permanent Notes/A Capacity Fix Sized to One Volume Driver Expires When the Driver Changes]] — CI sized for headcount growth, then met with AI-assisted commit volume — record the mix, not just the total. - [[3. Resources/Permanent Notes/Reclassify What Blocks Before Optimizing What Blocks]] — remove work from the bottleneck rather than speeding up the work already in it. ## Labor market & economics - [[3. Resources/Permanent Notes/Value-Per-Human Is the Real Mechanism Behind Replacing Employees With AI]] — the headless-business operating metric - [[3. Resources/Permanent Notes/Perm · The middle is disappearing as AI commodifies previously-expert work]] — barbell economy: bespoke experts + cheap commodity - [[3. Resources/Permanent Notes/AI-Driven Reset Is the Acceptable Public Frame for Margin-Driven Layoffs]] — "AI reset" label pollutes the labor signal (Elementor case) - [[3. Resources/Permanent Notes/A Cynical AI Bet Only Needs Bosses to Believe]] — layoffs can track investor incentives, not tool performance - [[3. Resources/Literature Notes/Lit · Why hasn't AI increased unemployment]] — Anthropic economics: adoption without displacement; pair with Doctorow - [[3. Resources/Literature Notes/Lit · AI solipsists and AI cynics]] — Doctorow's incentive account of the bubble; the counterweight - [[3. Resources/Permanent Notes/Perm · The Anthropic saga illustrates AI's geopolitical fragility]] — lab org-stability as the constraining variable - [[3. Resources/Literature Notes/Lit · Scaling a business with AI instead of employees]] — the headless-business pitch itself - [[3. Resources/Permanent Notes/Template-Presence Is the Diagnostic of Automatability]] — template-presence predicts AI exposure - [[3. Resources/Permanent Notes/Zero Build Cost Shifts Advantage From Depth to Breadth]] — cheap building → breadth beats depth - [[3. Resources/Permanent Notes/Abstracting Drudgery Re-Opens Creative Work to Non-Specialists]] — the abstraction→respecialization cycle - [[3. Resources/Permanent Notes/AI Buyers and AI Coworkers Are Opposite Design Postures]] — two agent roles, opposite UX - [[3. Resources/Permanent Notes/An Ops Agent Is Only as Capable as Its System Access]] — capability = access to systems of record - [[63% of Developers Engage With AI-Assisted Development]] — adoption baseline - [[3. Resources/Literature Notes/NotebookLM's New Vertical Shorts Are INCREDIBLE]] — a formerly gated feature moved to free as the underlying model got cheap enough to give away. - [[3. Resources/Permanent Notes/A Cheap Enough Model Turns a Gated Feature Into a Free One]] — once per-output cost drops far enough, marginal cost stops being the reason a vendor gates a feature behind a paywall. - [[3. Resources/Literature Notes/You Ask, I Answer Measuring AI ROI Beyond Time Saved]] — Penn/Trust Insights on measuring AI ROI, linked to existing baseline and vanity-metric permanents by the same author. - [[3. Resources/Permanent Notes/AI ROI Requires a Pre-Existing Performance Baseline]] — you can't quantify what AI changed if you never measured your own performance before it arrived. - [[3. Resources/Permanent Notes/Uplift Modeling Isolates a Treatment Effect From a Tracked Baseline]] — comparing a control period against a treatment period is what attributes a measured delta to the new input rather than to other confounders. - [[3. Resources/Permanent Notes/A Career Track Is a Reversible Seat Not an Identity]] — the IC/leadership move as an allocation of finite career time, with re-entry cost as the real constraint. ## Content & publishing under cheap generation - [[3. Resources/Permanent Notes/Perm · Content publishing at scale now requires automated curation filters, not just volume]] — curation > volume in the flooded landscape - [[3. Resources/Permanent Notes/Perm · AI-generated content slop becomes SEO-weaponized when low-effort production is cheap]] — cheap production as retrieval attack - [[3. Resources/Permanent Notes/Perm · The ecosystem will bifurcate between 'human-authored' and AI-scale producers]] — "human" becomes the premium label - [[3. Resources/Permanent Notes/Scaled Content Is Defined by Its Production Method Not Its Quality]] — Google's policy triggers on means of production - [[3. Resources/Permanent Notes/Shared AI Tools Homogenize Output While Feeling Original to Each User]] — convergence invisible individually, visible collectively - [[3. Resources/Permanent Notes/AI Design Tools Converge on the Tool-Maker's Own House Style]] — "Claudian sameness"; tools leave tells - [[3. Resources/Literature Notes/So What How AI Detectors Work]] — detectors measure word-order predictability, not authorship, and misfire hardest against non-native and neurodivergent writers - [[3. Resources/Permanent Notes/AI Editing Toward the Probable Is What Detection Detects]] — the Yoda minimal pair: reordering a human's own words into a statistically probable pattern is what triggers detection, not who wrote them - [[Concentration Beats Coverage Under Retrieval]] — retrieval rewards concentrated authority - [[Retrieval Quality Is Set by the Unit of Knowledge Not the Retrieval Algorithm]] — RAG accuracy is fixed upstream by the embedded unit - [[3. Resources/Literature Notes/My AI-Assisted Content System That Keeps Me in the Loop with Kyle Balmer]] — human-edit diffs as training signal - [[3. Resources/Literature Notes/You Ask, I Answer — Replacing Fine Tuned Models With RAG]] — Penn: hybrid RAG over fine-tuning; permission inheritance - [[AI for Content Summarization and Filtering]] · [[Almost Timely News - What ChatGPT Is Really Good At]] · [[Astonishingly Easy Peer-Review Writing - Uncover the Magic With ChatGPT!]] — what LLMs are genuinely good at - [[3. Resources/Literature Notes/Garofalo Details How AI Writes WordPress Social Content, Open Sources the Toolkit]] — WordPress.org's two-year AI social-content practice; both candidate claims already captured by the existing values-explicit-prompting and automation-removal permanents from this same source. - [[3. Resources/Literature Notes/The A.I.-Design Aesthetic That's Taking Over the Internet the New Yorker]] — companion piece to the existing house-style-convergence permanent, covering distinctiveness-as-labor rather than the convergence mechanism itself. - [[3. Resources/Permanent Notes/Distinctiveness Is the Labor an Opinionated AI Default Externalizes]] — once a generative tool ships an opinionated default, looking distinctive stops being free and becomes deliberate human work layered on top of the machine's output. - [[3. Resources/Permanent Notes/Emit Content Machines Can Consume Rather Than Embedding AI]] — the emitted surface outlives the model generation an embedded feature was built against. - [[3. Resources/Permanent Notes/A Format Whose Source Is Its Presentation Cannot Hide Bloat]] — visibility, not simplicity, is what keeps a format honest — and it breaks once tooling writes the source. ## WordPress & CMS disruption - [[3. Resources/Permanent Notes/Perm · Vibecoding bypasses CMS entirely, draining relative share to 'None']] — category bypass, not competition - [[3. Resources/Permanent Notes/WordPress's 2026 Losses Went to No-CMS Sites Not Competitors]] — the bake-off disappearing is the threat (Koenig) - [[3. Resources/Permanent Notes/Perm · Platform maturity is a lagging indicator on the S-curve]] — installed-base share hides weakening momentum - [[3. Resources/Permanent Notes/Perm · AI-powered content + design inside the editor closes the friction loop for user-generated velocity]] — fastest thought-to-deploy loop wins - [[3. Resources/Permanent Notes/AI Erodes the WordPress Plugin Value Props Pushing Monetization Toward SaaS and Services]] — value migrates to SaaS/services - [[3. Resources/Permanent Notes/Vernacular Software Survives the Bubble Because It Is Unmonetizable by Design]] — Doctorow: homemade AI code endures; tension with SDD - [[3. Resources/Permanent Notes/Cheap Adaptation Beats Good Prediction When Standards Multiply]] — make adaptation cheap instead of picking winners - [[3. Resources/Literature Notes/Build anything with Claude & WordPress]] — Ginder: direction-over-code, typing→steering - [[3. Resources/Literature Notes/Create Your First WordPress Plugin Using Cursor (Complete Tutorial)]] — one prompt → working plugin economics - [[3. Resources/Literature Notes/I Built a WordPress Plugin Nobody Asked For — Here's Why and How]] — skills collapse personal-tool build cost - [[3. Resources/Literature Notes/AI Fatigue Is an Opportunity for Agencies]] — client AI-fatigue as services pipeline - [[3. Resources/Literature Notes/The Modern WooCommerce Developer Workflow]] — build on the stable extension layer - [[3. Resources/Literature Notes/Lit · Centralized AI management improves user experience]] · [[3. Resources/Literature Notes/Lit · MCP protocol enables conversational WordPress publishing]] · [[3. Resources/Literature Notes/Lit · Mistral integration demonstrates faster LLM performance]] · [[3. Resources/Literature Notes/Lit · Open source AI models closing quality gap rapidly]] — WordPress AI-integration tracking - [[3. Resources/Literature Notes/WordPress Drops to 41.5% Market Share]] — full deep-read whose three candidate claims (lagging indicator, no-CMS losses, vibecoding-drains-share) were already promoted in a prior pass; cross-linked to all three existing permanents. - [[3. Resources/Literature Notes/An Agent-First Approach to Create Block Theme - Brian Coords]] — an agent-first WordPress tooling case study, ported directly from an existing full deep-read. - [[3. Resources/Permanent Notes/Agent-First Tooling Exposes Capability Through a CLI Not a New Plugin]] — a CLI wrapper around an existing internal capability, not a new GUI plugin, is the fastest way to make it usable by a coding agent. - [[3. Resources/Literature Notes/Studio - The Future of AI WordPress Is Now]] — The WP Minute's walkthrough of Automattic Studio 2.0 beta's agentic WordPress builder. - [[3. Resources/Permanent Notes/An Agentic Builders Safety Signal Is Convention Adherence Not Output Polish]] — whether an agentic WordPress builder is safe to hand a client depends more on whether it follows platform-native conventions than on how polished its output looks. ## Maintenance & durability - [[3. Resources/Permanent Notes/AI Coding Defaults to Addition Over Maintenance]] — immediate prompt satisfaction favors duplicated additions while shared abstractions and legacy code decay. - [[3. Resources/Permanent Notes/A Stale Backlog Creates Shadow Systems]] — shared systems lose authority when abandoned items make private tracking more reliable. - [[3. Resources/Permanent Notes/Cheap Generation Raises the Total Maintenance Bill]] — generation is the smallest cost the artifact will ever have - [[3. Resources/Readwise/Articles/Maintaining Over Building]] — Coords: 90% of vibe-coded output abandoned; the origin source - [[3. Resources/Permanent Notes/Perm · Automation removal reveals its own value via growth drop]] — natural-experiment methodology for automation ROI - [[3. Resources/Readwise/Articles/A Jazzy Organisation]] — improvisation-over-process when coordination is the constraint - [[3. Resources/Literature Notes/WordPress Agencies Aren't Afraid of AI. They're Buried in Updates]] — survey evidence that maintenance load, not AI hesitancy, is the actual agency-side bottleneck; its adoption-inverted-to-pain claim links to an existing permanent from the same source. - [[3. Resources/Permanent Notes/An Agency's Maintenance Load Tracks Site Count Not Headcount]] — a small team can carry portfolio-scale maintenance risk because the real burden scales with sites managed, not staff count. ## PKM & formation - [[3. Resources/Permanent Notes/AI-Assisted PKM Replaces Synthesis Work With the Appearance of Synthesis Work]] — AI substituting for the formative work - [[3. Resources/Permanent Notes/AI Literacy Is Relational Not Just Technical]] — trust, dependence, and repair, not just prompting - [[3. Resources/Permanent Notes/AI Is a Building Block You Build From Not a Tool You Add to a Workflow]] — subtraction test: does the workflow make sense without it? - [[3. Resources/Permanent Notes/Notebooks Are JSON So Agentic Work Becomes an Auditable Memory Artifact]] — the record as retrospective memory - [[3. Resources/Literature Notes/Claude Cowork Finally Clicked When I Learned This]] — folder-scoped instruction files end re-explaining - [[3. Resources/Permanent Notes/A Diagnostic Quiz Before Content Generation Tailors the Learning Path]] — a pre-generation diagnostic quiz turns a generic lesson plan into a tailored one and makes progress tracking measure something real. ## Key sources - [[3. Resources/Readwise/Articles/Canaries in the Coal Mine Six Facts About the Recent Employment Effects of Artificial Intelligence]] — the ADP payroll data behind the −13% - [Knowledge-Worker Productivity...](3.%20Resources/Readwise/Articles/Knowledge-Worker%20Productivity....md) — Drucker's Taylor move, the historical template - [[3. Resources/Readwise/Articles/The Marks are on the Knowledge Worker]] — output variation as the knowledge worker's value - [[3. Resources/Readwise/Articles/Holy Sht I Think Anthropic Is Profitable Now]] — the pricing/economics backdrop - [[3. Resources/Readwise/Snipd/Data/Podcast – Cory Doctorow's craphound.com/The Reverse Centaur’s Guide to Life After AI]] — the centaur/reverse-centaur primary source - [[3. Resources/Literature Notes/Anthropic's New "Research" Report is Dumb]] · [[3. Resources/Literature Notes/The Next Horizon in Agents]] · [[3. Resources/Literature Notes/Anthropic Introduces Claude for Teachers]] · [[3. Resources/Literature Notes/Features of GitHub Copilot]] — vendor-claim media literacy + agent-market landscape ## Idea Compass - **Parents**: labor economics of automation · Drucker on knowledge-worker productivity · apprenticeship and tacit-knowledge transmission - **Children**: DevRel pipeline under compression · maintainer pipeline under compression · vendor default-on as reverse-centaur push - **Friends**: [[3. Resources/Permanent Notes/AI-Assisted PKM Replaces Synthesis Work With the Appearance of Synthesis Work]] · [[3. Resources/Permanent Notes/Stability Is the Precondition for Formation Across Domains]] - **Competitors**: pure-augmentation thesis (pipeline self-heals) · senior hiring from non-traditional paths (apprenticeship-free) ## Open Questions to Track - Early empirical signals the senior-judgment pipeline is breaking (senior tenure, age distribution, time-to-promotion)? - Documented cases of AI adoption *with* deliberate apprenticeship investment? - Does compression apply equally to creative, analytical, and relational work? - DevRel: if "AI does the demo" is the compressed entry work, what produces senior DevRel judgment? ## Related MOCs - [[AI Agent Safety MOC]] — operational controls side of the same deployments - [[AI Cost & Tokenomics MOC]] — the money side; review-tax and token spend as cost-shadow - [[DevRel MOC]] — where pipeline compression is most concrete here - [[Open Source Governance MOC]] — maintainer-attention scarcity as the OSS instance - [[Personal Knowledge Management MOC]] — the synthesis-substitution analog ## Synthesis additions (2026-08-01, run 2) - [[3. Resources/Permanent Notes/Fluency Makes Small Errors Harder to Catch Than Large Ones]] — why review attention goes to the wrong artifacts. Generation severed the fluency-competence correlation, so the polished paragraph deserves the scrutiny the messy one appears to. Argues for preferring executable outputs, where fluency buys nothing with a machine. - [[3. Resources/Permanent Notes/Automation Relocates Toil Into Supervision]] — the unbudgeted cost of agentic workflows. Carries `proof_state: working` and a falsifier: a before/after time-and-motion study that counts review and prompting as work. Most published productivity claims do not. - [[3. Resources/Permanent Notes/Silent Failures Cost More Than Loud Ones]] — the cost curve behind the fluency problem: detection time dominates repair time, so review budgets should follow *detectability* rather than apparent complexity. Its scope names the real exception — where the loud failure is catastrophic and irreversible, crashing later is worse than being quietly wrong now. ## Synthesis additions (2026-08-01, run 3–4) - [[3. Resources/Permanent Notes/Prevented Work Is Invisible to Velocity Metrics]] — the measurement failure underneath AI-era burnout. Labor that stops a defect from existing produces no artifact, so throughput rises while the people absorbing the review load register as the least productive on the dashboard. - [[3. Resources/Permanent Notes/AI Adoption Concentrates Where It Is Pleasant Not Where the Pain Is]] — why near-universal adoption can leave the actual bottleneck untouched. The WordPress agency survey is the cleanest evidence: ~90% AI adoption, 16% for maintenance, and maintenance is the top-ranked pain. - [[3. Resources/Permanent Notes/Context Not Constraints Is What Unhobbles an Agent]] — a mechanism for why the diagnosis order (context → effort → model) puts context first, with a measurement attached: ~80% of the Claude Code system prompt removed, product improved. ## Synthesis additions (2026-08-01, run 5) - [[3. Resources/Permanent Notes/Watching a System Run Beats Reading It]] — the discipline that matters most where generated explanations are fluent and cheap. A model's account of what code does is a reading, and readings are exactly what generation produces plausibly. ## Synthesis additions (2026-08-08) - [[3. Resources/Permanent Notes/Adversarial Practice Builds Transferable Confidence Through Worst-Case Exposure]] — orphan fix: the note itself already pointed here ("See also"), but the link was one-directional. Confidence that survives a real high-stakes interaction comes from practicing against the worst version of it, not the encouraging average a sycophantic AI tool defaults to — a concrete case of AI-tool design choices (which model, how tuned) determining whether the practice is useful or just comfortable. ## Synthesis additions (2026-08-11, backlog clearance) - [[3. Resources/Readwise/Articles/Our Vibe Coded Projects That Actually Work the Vergecast]] — supports the existing judgment-bottleneck and maintenance-cost claims with practitioner evidence about throwaway builds, weak security literacy, and idea scarcity. - [[3. Resources/Readwise/Articles/Field Guide to Fable — Thariq Shihipar, Anthropic]] — first-party account of capability overhang and finding unknowns, retained with its promotional and experiential limits. - [So I've Been Using GPT-5.6 for Awhile...](3.%20Resources/Readwise/Articles/So%20I've%20Been%20Using%20GPT-5.6%20for%20Awhile....md) — practitioner model assessment retained as time-bound workflow evidence rather than a durable capability claim. ## Synthesis pass — 2026-08-11 WordPress reference - [SiteGround WordPress small-business AI report](https://read.readwise.io/read/01ky8e32qcryw2mzmtc0yxmnc3) — vendor survey evidence that frequent AI use can coexist with persistent quality friction; retained as a source rather than generalized into a new claim. ## Synthesis additions (2026-08-16 — Goodhart's Law and metrics gaming) - [[3. Resources/Literature Notes/The 8 Software Engineering Metrics AI Broke]] — companion source to the existing activity-metrics note, isolating the Goodhart's-Law mechanism as its own claim. - [[3. Resources/Permanent Notes/Goodharts Law Becomes Automatic Under AI Tool Use]] — AI's speed turns any chosen metric into a target that gets gamed without anyone deciding to game it. - [[3. Resources/Literature Notes/Why Instructional Designers Should Learn to Challenge AI]] — instructional-design case for building a pressure stage against AI output instead of trusting first drafts. - [[3. Resources/Permanent Notes/AI Workflows Need a Pressure Stage Not Just a Generation Stage]] — a workflow that only generates and never stress-tests inherits Goodhart's failure mode by default. - [[3. Resources/Literature Notes/AI-coding agents kill team collaboration]] — self-review-by-the-same-tool collapsing team collaboration, a sibling mechanism to the existing team-capability-aggregation note. - [[3. Resources/Permanent Notes/Tool Access Does Not Aggregate Into Team Capability]] — individual agent proficiency doesn't sum into organizational capability without shared review. ## Synthesis additions (2026-08-16 — harness, context, and agent architecture) - [[3. Resources/Literature Notes/How to Use Claude AI for Coding (Boris Cherny)]] — practitioner account of feedback-loop design as the lever on agent coding quality. - [[3. Resources/Permanent Notes/Agent Output Quality Scales With the Feedback Loop Not the Prompt]] — the loop that checks output matters more than how the prompt was worded. - [[3. Resources/Literature Notes/Cloudflare Acquires VoidZero]] — infrastructure consolidation as a precondition for agent legibility, not just a build-tooling story. - [[3. Resources/Permanent Notes/Platform Consolidation Makes a Stack Legible to AI Agents]] — fewer, unified tools give an agent one coherent surface to reason about instead of many partial ones. - [[3. Resources/Literature Notes/The Reason AI Gives You Useless Answers (Fix This First)]] — Forte's PCM framework locating bad output in missing staged context, not prompt wording. - [[3. Resources/Permanent Notes/AI Output Quality Is Bounded by Staged Context Not Prompt Wording]] — companion to the existing vague-prompts and context-not-constraints notes; context ceiling, not phrasing, sets the quality ceiling. - [[3. Resources/Literature Notes/Technical debt is becoming an AI problem.md|Technical debt is becoming an AI problem]] — internal discoverability, not model capability, blocks agents from safely touching legacy systems. - [[3. Resources/Permanent Notes/An Organization Cannot Expose Data to a New System It Cannot Locate Internally]] — you can't grant an agent access to data your own org can't find. - [[3. Resources/Literature Notes/Project Fetch, Phase two (Anthropic)]] — supporting evidence for the existing three-stage capability-arc note, same claim from a later checkpoint. - [[3. Resources/Literature Notes/Firecrawl MCP Server Search Scrape And Crawl The Web]] — bad web answers traced to fetch-tooling gaps rather than model reasoning failures. - [[3. Resources/Permanent Notes/Wrong Web Answers Trace to Fetch Tooling Not the Model]] — the harness-over-model principle applied specifically to live web retrieval. - [[3. Resources/Literature Notes/Cloudflare Is Quietly Building the Agent Cloud]] — argues an agent is a durable distributed workflow, not a stateless chatbot loop. - [[3. Resources/Permanent Notes/Model An Agent As A Durable Workflow Not A Chatbot]] — architecture implication: design for persistence and recovery, not just a single conversational turn. - [[3. Resources/Literature Notes/Design Tokens Manual Setup and AI Workflow (Penpot)]] — a design tool's exposed API sets the ceiling on what an agent can actually edit. - [[3. Resources/Permanent Notes/An Agent's Design-Editing Power Is Bounded by the Tool's Exposed API]] — the tool's surface, not the model, caps agentic design editing. - [[3. Resources/Literature Notes/Field Guide to Fable — Thariq Shihipar, Anthropic]] — fuller Lit Note treatment of the source already cited in the 2026-08-11 backlog clearance. - [[3. Resources/Permanent Notes/Agents Shift Human Work To Specifying Intent]] — as execution gets delegated, the human's job moves to naming what's wanted precisely. - [[3. Resources/Literature Notes/A Field Guide to Fable Finding Your Unknowns]] — the map/territory framing for AI prompts, combined from two quoted atoms per atomicity rule. - [[3. Resources/Permanent Notes/AI Prompts Are Maps Not the Territory]] — a prompt is a compressed representation, not the thing itself; treat drift between them as expected. - [[3. Resources/Literature Notes/AI Agents in GTM What They Should Truly Do]] — GTM-specific case for naming the job before picking the automation. - [[3. Resources/Permanent Notes/Define the Job Before Selecting the Automation]] — automation choice should follow a defined job, not precede it. - [[3. Resources/Literature Notes/Welcome to rss.chat (Dave Winer)]] — in-app AI collaboration outperforming an external consulting relationship. - [[3. Resources/Permanent Notes/AI Works Better as an In-App Collaborator Than an External Consultant]] — proximity to the work beats general-purpose external advice. - [[3. Resources/Literature Notes/Can MCP Servers Actually Reduce Streaming Workload]] — a promotional source with a genuine claim about where MCP currently adds value. - [[3. Resources/Permanent Notes/MCP Servers Add the Most Value on Asynchronous Existing Actions]] — MCP's current scope is narrower than the hype; it shines on async work against existing actions. - [[3. Resources/Literature Notes/I can't believe they released this]] — orchestration termination logic as a distinct control problem from model choice. - [[3. Resources/Permanent Notes/Orchestration That Terminates by Code Beats Termination by Model Choice]] — deterministic stop conditions in code outperform trusting the model to know when it's done. - [[3. Resources/Literature Notes/Almost Timely News - How to Connect an AI Agent to a Data Source]] — connects an agent to a data source while distinguishing planning from execution scale. - [[3. Resources/Permanent Notes/An Agent Is Defined by Working Without Mid-Task Steering]] — the definitional line between "agent" and "assistant" is whether a human steers mid-task. - [[3. Resources/Literature Notes/I Built a Headless WooCommerce Store Using AI + Terminal]] — vendor demo with a genuinely reusable debugging mechanism. - [[3. Resources/Permanent Notes/Give an AI Agent the Raw Error Not Your Paraphrase]] — paraphrasing an error strips the exact signal an agent needs to diagnose it. - [[3. Resources/Literature Notes/How I use Agents to Extract Hidden Feedback from Github and Improved Rector]] — narrowing a prompt once the task is named beats broad exploratory prompting. - [[3. Resources/Permanent Notes/Narrow Agent Prompts Beat Broad Exploratory Ones Once the Task Is Named]] — specificity pays off once you know what you're actually asking for. - [[3. Resources/Literature Notes/Taylor Otwell on Why Laravel + AI Pair So Well, and the Future of Programming]] — a framework's conventions function as free, always-available context for an agent. - [[3. Resources/Permanent Notes/Framework Conventions Function as Executable Context for Agents]] — well-established conventions substitute for context an agent would otherwise need spelled out. - [[3. Resources/Literature Notes/How We Teach AI Models]] — training-layer breakdown (pretrain/SFT/RL/eval) motivating written specs as artifacts. - [[3. Resources/Permanent Notes/A Written Model Specification Turns Desired Behavior Into an Inspectable Artifact]] — a written spec makes "desired behavior" something you can check, not just intuit. - [[3. Resources/Literature Notes/How AI Agents Work Prompt, Context, Harness and Loop Explained]] — a taxonomy separating the four control surfaces of an agent system. - [[3. Resources/Permanent Notes/Prompt Context Harness and Loop Are Separate Agent Control Problems]] — each of prompt, context, harness, and loop fails independently and needs its own fix. - [[3. Resources/Literature Notes/The New Rules of Context Engineering for Claude 5 Models]] — more capable models need fewer, not more, guardrail constraints. - [[3. Resources/Permanent Notes/Capable Models Need Fewer Constraints Not More]] — over-constraining a capable model costs more than it protects. - [[3. Resources/Literature Notes/Liberating sites with Studio Code (Bossenger)]] — migration agent UX case for step-level progress reporting. - [[3. Resources/Permanent Notes/AI Migration Agents Should Report Step-Level Progress]] — users trust long-running agent migrations more when progress is visible step by step. - [[3. Resources/Literature Notes/Anthropic Just Revealed How to Prompt Opus 5]] — official prompting guidance for Opus 5, cross-checked against independently-arrived-at claims elsewhere in this MOC. - [[3. Resources/Literature Notes/Why AI is going vertical (again) - Dianne Penn (Anthropic)]] — vertical AI product strategy grounded in evals as a requirements-writing discipline. - [[3. Resources/Permanent Notes/Evals Function as Executable Requirements Not PRD Replacements]] — writing an eval forces the same precision a PRD claims to provide, but it's checkable. - [[3. Resources/Literature Notes/Fable Broke My App and Couldn't Fix It]] — when diagnosis fails, reframing the agent's task toward tool-building recovers value. - [[3. Resources/Permanent Notes/Reframing an Agent's Task From Diagnosis to Tool-Building Recovers Value After Diagnosis Fails]] — a failed diagnostic run isn't wasted if it's redirected into building the tool that would diagnose it. - [[3. Resources/Literature Notes/Navigating Claude Code - The Bare Minimum Setup]] — series entry 2/10; prose config isn't a real access boundary. - [[3. Resources/Permanent Notes/Prose Configuration Cannot Substitute for a Real Access Boundary]] — deny rules written in prose are guidance, not enforcement. - [[3. Resources/Literature Notes/Navigating Claude Code - CLAUDE.md Done Right]] — series entry 3/10; durable-context files decay without active pruning. - [[3. Resources/Permanent Notes/A Durable-Context File Needs Active Pruning Not Just Careful Writing]] — a CLAUDE.md written well once still rots without ongoing maintenance. - [[3. Resources/Literature Notes/Navigating Claude Code - The Context Window Tax]] — series entry 4/10; connects to the existing "declared capability costs every turn" pattern rather than duplicating it. - [[3. Resources/Literature Notes/Navigating Claude Code - Skills That Actually Fire]] — series entry 10/10; a skill's description decides whether it fires, independent of instruction quality. - [[3. Resources/Permanent Notes/A Skills Description Decides Whether Its Used Independent of Instruction Quality]] — the invocation trigger and the instruction body are two separate design problems. - [[3. Resources/Literature Notes/Penpot MCP vs Figma MCP The Real Estate Card Test]] — comparative test of agent-readable vs opaque-canvas design files. - [[3. Resources/Permanent Notes/Agent-Readable Design Files Beat Opaque Canvases for AI Output]] — structured, readable design data outperforms a rendered canvas as agent input. - [[3. Resources/Literature Notes/Penpot AI Kit — Agent Skills for Your Design Files (2026)]] — a skills repository functioning as the client library for an MCP server. - [[3. Resources/Permanent Notes/A Skills Repository Is the Client Library for an MCP Server]] — skills packages are the ergonomic wrapper around a raw MCP surface, same role a client library plays for an API. - [[3. Resources/Literature Notes/Using WordPress Studio Code to Build a Hockey Team Site]] — AI coding agents doing well on mechanical bulk work, with a Gutenberg 7.1 breaking-change warning surfaced along the way. - [[3. Resources/Permanent Notes/AI Coding Agents Excel at Mechanical Bulk Work]] — the current strength is repetitive, well-specified work — not novel judgment calls. - [[3. Resources/Literature Notes/I Made My AI Agent Fix Its Own Copy-Paste (jscpd + Mago)]] — directly actionable for Courtney's own maintained plugin repos. - [[3. Resources/Permanent Notes/Automated Guard Rails Should Match Mechanical Verifiability]] — automate the checks that are mechanically checkable; leave judgment calls to review. - [[3. Resources/Literature Notes/Almost Timely News How To Expand and Improve Content with AI, Part 1]] — gap-analysis prompting treated as a distinct skill from generation prompting. - [[3. Resources/Permanent Notes/Gap-Analysis Prompting Is a Distinct Skill From Generation Prompting]] — finding what's missing requires a different prompt discipline than producing more of what's there. - [[3. Resources/Literature Notes/You Ask, I Answer - Handling AI Context Drift]] — includes a disclosed prompt-injection-shaped payload in the source, recorded as an artifact and not acted on. - [[3. Resources/Permanent Notes/A Frozen Knowledge Base Entrenches Whoever Was Prominent at Cutoff]] — a static knowledge base locks in the prominence rankings of its training-cutoff moment. - [[3. Resources/Permanent Notes/A Stuck State Needs a Reset Not a Memory Fix]] — salvages the source's correct reasoning pattern even though its own stated conclusion was wrong. - [[3. Resources/Literature Notes/WordPress 7.1 is afoot, Gutenberg 23.6, iframed post editor, Abilities API, Studio Code and more - Weekend Edition 371]] — release roundup surfacing both a deprecation deadline and a corpus-bias pattern. - [[3. Resources/Permanent Notes/An Agent Defaults to the Corpus-Prevalent Pattern Not the Currently Recommended One]] — an agent trained on old examples reaches for the common pattern over the newly-recommended one. - [[3. Resources/Literature Notes/226 – Jessica Lyschik on Why Accessibility in WordPress Themes Is Easier Than You Think]] — linked for its machine-consumption argument, not its accessibility-standards content. - [[3. Resources/Permanent Notes/Machine Consumers Make Semantic Markup a Commercial Requirement]] — AI agents parsing markup turn semantic correctness into a business requirement, not just a courtesy. - [[3. Resources/Literature Notes/AI Go-to-Market Strategy (Common Room)]] — both candidate claims already existed as permanent notes elsewhere; linked rather than duplicated. - [[3. Resources/Literature Notes/AI News NotebookLM Gemini Study Notebooks Claude Design 2.0]] — rumor-heavy roundup; only the confidently-stated claim was promoted. - [[3. Resources/Permanent Notes/In-Context AI Assistants Lower the Adoption Barrier More Than Standalone Apps]] — meeting users where they already are beats asking them to open a new app. - [[3. Resources/Literature Notes/how did design fix the ai design slop problem]] — companion evidence for the existing policy-gap-not-capability-gap note. - [[3. Resources/Literature Notes/Keep Claude Code Up to Date on Time and Date (Tom McFarlin)]] — practical case for the existing volatile-context-at-the-event-where-it-changes note. ## Synthesis additions (2026-08-16 — judgment, review, and verification) - [[3. Resources/Literature Notes/Marquee Gallery becomes a Content Gallery - Prompt Library Angie]] — a curated prompt library treated as a starting point, not an answer key. - [[3. Resources/Permanent Notes/A Vetted Prompt Is a Starting Point to Refine Not a Finished Answer]] — even a vetted prompt still needs situational refinement before use. - [[3. Resources/Literature Notes/More Targeted Content Editing With Block MCP]] — AI editing tool re-renders the real platform rather than a simulated preview. - [[3. Resources/Permanent Notes/AI Editing Tools Earn Trust By Re-Rendering The Real Platform]] — trust comes from validating against the actual render target, not a proxy. - [[3. Resources/Literature Notes/You Can Vibe Code a Demo. You Can't Vibe Code a Product]] — the demo-to-product gap as a companion to the existing implementation-vs-taste bottleneck note. - [[3. Resources/Permanent Notes/The Demo-to-Product Gap Is Human Work Not Model Work]] — closing the gap between a working demo and a shippable product is judgment work no model does for you. - [[3. Resources/Literature Notes/Revisiting What Matters Most]] — a rich source yielding two promoted claims on uncertainty and incrementalism. - [[3. Resources/Permanent Notes/Uncertainty Expression Is a Shipping Gate for AI Tools]] — a tool that can't express its own uncertainty shouldn't ship a confident-sounding answer. - [[3. Resources/Literature Notes/When Scrutiny Becomes Cheap]] — paired with the existing bubbles-end-from-cost note as two cost-collapse stories on opposite sides of the same shift. - [[3. Resources/Permanent Notes/When an Inspection Cost Collapses Signal-Based Value Reprices]] — once checking is nearly free, things that traded on the appearance of quality lose value fast. - [[3. Resources/Literature Notes/It Is Trivially Easy to Use Reddit to Manipulate AI Search]] — backfilled Lit Note for the existing AI-search-inherits-trust-and-fragility permanent. - [[3. Resources/Literature Notes/Intuition Is the Differentiator (Harper Carroll)]] — thin source, single combined claim about intuition as the automation-resistant skill. - [[3. Resources/Permanent Notes/Intuition Is the Automation-Resistant Axis of Human Skill]] — the part of expertise hardest to automate is the part that isn't explicit reasoning. - [[3. Resources/Literature Notes/OpenAI Codex Lead on the New Shape of Product Work (Andrew Ambrosino)]] — connects to the existing implementation-is-cheap-taste-is-the-bottleneck note; one distinct new claim promoted. - [[3. Resources/Permanent Notes/A Prototype's Polish Signals Nothing About Its Process Stage]] — a polished-looking prototype can be at any stage of the process; polish isn't a stage signal. - [[3. Resources/Literature Notes/WordPress Studio MCP (Jonathan Bossenger)]] — disambiguated by author suffix from a different WordPress Studio MCP video. - [[3. Resources/Permanent Notes/Agentic Tooling Shifts the Developer From Author to Editor]] — the companion to "the bottleneck moved to review": the developer's own role shifts from writing to editing. - [[3. Resources/Literature Notes/I Don't Know Rust. My AI Is Rewriting PHP in It Anyway|I Don't Know Rust. My AI Is Rewriting PHP in It Anyway.]] — two distinct trust-and-failure-mode claims from one rich deep-read. - [[3. Resources/Permanent Notes/External Test Oracles Make AI-Generated Code Trustworthy]] — trust in AI-generated code comes from an external oracle checking it, not from the code's fluency. - [[3. Resources/Permanent Notes/AI-Generated Code Hides Silently-Broken Potemkin Features]] — a feature can look complete and pass a glance while being silently non-functional underneath. - [[3. Resources/Literature Notes/So I've been using gpt-5.6 for awhile]] — manual-fallback transcript with no live Readwise highlights, treated as unverified anecdote. - [[3. Resources/Permanent Notes/Vendor Usage Anecdotes Need Verification Before Adoption Decisions]] — one practitioner's informal impression of a model isn't evidence for an adoption decision. - [[3. Resources/Permanent Notes/A Vendors Capability Claim Needs a Paired Comparison Not a Demo]] — a vendor demo alone doesn't establish capability; you need a comparison against the incumbent. - [[3. Resources/Literature Notes/Stop Reading Every Line of Code]] — backfilled Lit Note for the existing disposable-code-produces-evidence permanent; one new distinct claim promoted. - [[3. Resources/Permanent Notes/Review Scrutiny Should Scale With Consequence Not Production Method]] — how a change was produced shouldn't set review depth; what it can break should. - [[3. Resources/Literature Notes/MCP 2026-07-28 the spec catches up with the static web]] — one candidate claim was a duplicate of an existing canonical-sources note and was linked instead of recreated. - [[3. Resources/Permanent Notes/Cheaper Mechanics Never Substitute for Editorial Judgement]] — making distribution mechanically cheaper doesn't replace the editorial judgment of what to distribute. - [[3. Resources/Literature Notes/Prompt Engineering Is No More - Strategic Thinking Has Returned]] — thin, unevidenced opinion piece; backfilled Lit Note for the existing fluency-conceals-small-errors permanent. - [[3. Resources/Literature Notes/Give Your Agents a Performance Review]] — argues an agent evaluation is meaningless without an incumbent baseline to compare against. - [[3. Resources/Permanent Notes/An Agent Evaluation Without an Incumbent Baseline Cannot Establish Value]] — "the agent did X" only means something relative to what the human/process it replaced could do. - [[3. Resources/Literature Notes/Cloudflare Wallets - A REAL React Compiler Released (Syntax Weekly)]] — fast-moving roundup; the "meat proxy" review-gap claim was the one genuinely new candidate. - [[3. Resources/Permanent Notes/Human Review Is the Actual Authenticity Control Not AI Use Itself]] — the control that matters is whether a human reviewed the output, not whether AI touched it. - [[3. Resources/Literature Notes/Joey Kudish - how I use AI these days]] — missing Lit Note for a source already referenced as a link target elsewhere in the vault. - [[3. Resources/Permanent Notes/Review the Plan Not Just the Output]] — reviewing the plan an agent will execute catches problems earlier than reviewing only what it produced. - [[3. Resources/Literature Notes/The Rise of the Design Engineer]] — links to the existing external-test-oracles note as an independently-convergent claim. - [[3. Resources/Permanent Notes/Cheap Generation Raises the Value of Deterministic Verification]] — the cheaper generation gets, the more a deterministic check is worth relative to it. - [[3. Resources/Literature Notes/Automattbot]] — very short source; one proportionate permanent note promoted. - [[3. Resources/Permanent Notes/A Persona Bot Reply Is Not the Person's Current Judgment]] — a branded persona bot's reply carries the brand's liability, not the named person's actual current opinion. ## Synthesis additions (2026-08-16 — apprenticeship, learning, and pairing) - [[3. Resources/Literature Notes/Coding Interviews in the AI Era - Beyond LeetCode for Career Growth]] — backfilled Lit Note; existing permanent already covered the interview-format claim, so only the distinct pairing claim was promoted. - [[3. Resources/Permanent Notes/Always-Available AI Pairing Decouples Output-Pairing From Social-Pairing]] — AI pair programming satisfies the output half of pairing while leaving the social/mentorship half unmet. - [[3. Resources/Literature Notes/ChatGPT and AI Literacy — How Students Learn Through Use]] — Ammari et al. 2026 study distillation on how AI use patterns shape student learning. - [[3. Resources/Permanent Notes/AI Literacy Is a Post-Answer Behavior Not a Property of the Answer]] — literacy shows up in what a student does after getting an AI answer, not in the answer itself. - [[3. Resources/Permanent Notes/Naming AI-Use Modes Turns Habit Into Deliberate Strategy]] — labeling how you're using AI (e.g. "brainstorm mode" vs "check mode") converts unconscious habit into chosen strategy. - [[3. Resources/Literature Notes/We Must Stop Using AI to 'Level-Down' Our Students]] — cross-linked to the existing raise-challenge-not-lower-expectations and policy-gap notes. - [[3. Resources/Permanent Notes/Automation That Absorbs the Effort Absorbs the Learning]] — effort and learning are coupled; automating away the effort removes the learning that rode along with it. - [[3. Resources/Permanent Notes/Scaffolds Change the Route Substitutes Change the Artifact]] — a scaffold changes how a student gets to the answer; a substitute changes what answer they turn in. - [[3. Resources/Literature Notes/Back and Forth — AI as a Communication Device in WordPress Contribution]] — Anne McCarthy's account of AI prototypes as a communication device inside WordPress core contribution. - [[3. Resources/Permanent Notes/AI Prototypes Communicate Intent Better Than Competing Pull Requests]] — a runnable prototype communicates a proposal's intent faster than a written PR competing for the same decision. - [[3. Resources/Permanent Notes/Working Locally With No-Push Intent Limits AI Contribution Risk]] — keeping AI-assisted contribution local and non-pushed bounds the blast radius of a bad suggestion. ## Synthesis additions (2026-08-16 — power, deployment, and burnout) - [[3. Resources/Literature Notes/AI Is Not a Tool]] — its two strongest claims were already captured by existing permanents; one genuinely new leadership-posture claim promoted. - [[3. Resources/Permanent Notes/AI-First Leadership Holds The Outcome Not The Path]] — AI-first leadership specifies the outcome wanted and lets the path to it vary, rather than dictating both. - [[3. Resources/Literature Notes/Rethinking Developer Life and Productivity with Rapid AI Advancements]] — backfilled Lit Note; both its burnout-adjacent claims were already captured by existing permanents, so only a distinct technical claim was newly promoted (filed under WordPress & CMS disruption below). - [[3. Resources/Literature Notes/Rethinking Developer Life and Productivity With Rapid AI Advancements]] — near-duplicate title of the entry above from an apparent concurrent-processing collision on the same source; kept because it produced a genuinely distinct permanent. - [[3. Resources/Permanent Notes/Reclaim AIs Saved Time Instead Of Reinvesting It]] — the response half of the fatigue-signal problem: time AI frees up gets reinvested into more output instead of reclaimed as slack. - [[3. Resources/Literature Notes/We Just Formed the Largest Tech Union in America]] — backfilled Lit Note for the existing AI-deployment-is-a-bargaining-issue permanent. - [[3. Resources/Permanent Notes/Job Security Is a Precondition for Innovation Not a Constraint]] — workers experiment more, not less, when their job isn't on the line for the experiment failing. - [[3. Resources/Literature Notes/Agentrification (Scott Logic)]] — its custodian-trap claim already existed; only the status-driven-adoption claim was new. - [[3. Resources/Permanent Notes/Tooling Chosen for Status Rather Than Utility Is a Cost]] — adopting a tool to signal status rather than for its utility still carries the tool's full switching and maintenance cost. - [[3. Resources/Literature Notes/Why AI Feels Overwhelming to Leaders]] — vendor-pitch source flagged for its unverified 90-day marketing claim; the underlying diagnosis is genuine. - [[3. Resources/Permanent Notes/Choice Overload Stalls Adoption in an Already-Motivated Population]] — too many AI options stalls even leaders who already want to adopt. - [[3. Resources/Literature Notes/Your AI Agents Have Dashboards. Their Human Managers Have Annual Reviews]] — asymmetry between real-time agent monitoring and slow human performance review cycles. - [[3. Resources/Permanent Notes/Delegating a Decision to AI Does Not Transfer Accountability for It]] — the human who delegated a decision to AI still owns its consequences. - [[3. Resources/Literature Notes/The AI Hypocrisy - Secretive AI Use and Creator Backlash (Penn)]] — cross-linked as a contrast (different affected population) and an extension of existing disclosure-power notes. - [[3. Resources/Permanent Notes/Concealment Converts One Grievance Into Two]] — hiding AI use turns "I used AI" into "I used AI and I lied about it" — a second, separate grievance. - [[3. Resources/Literature Notes/Just Trying to Figure Out My Job]] — Hank Green's own primary account, cross-linked to a secondhand analysis of the same incident elsewhere in the vault. - [[3. Resources/Permanent Notes/Accepting Valid Criticism Does Not Require Accepting Every Claim Within It]] — you can concede the part of a criticism that's true without conceding the parts that aren't. - [[3. Resources/Literature Notes/AI and the Enshittification Era (Cory Doctorow)]] — richest source in this batch; three of five candidate claims promoted. - [[3. Resources/Permanent Notes/Automation Financed by Capital Paces Workers Labor Assists Them]] — who pays for the automation determines whether it sets the pace for workers or assists them at their own pace. - [[3. Resources/Permanent Notes/Replacing a Worker Destroys Process Knowledge That Rehiring Cannot Restore]] — the tacit knowledge a laid-off worker held doesn't come back when you rehire for the role. - [[3. Resources/Permanent Notes/A Single Human Supervising Automation Absorbs Liability the System Created]] — the "human in the loop" often exists to hold liability the automated system generated, not to add judgment. - [[3. Resources/Literature Notes/The CEO Goes First]] — the incentive half of an adoption programme, paired with a sibling clip covering the enforcement half. - [[3. Resources/Permanent Notes/The Uncommitted Middle Judges Adoption by Leadership Behavior Not Announcements]] — undecided employees watch what leaders actually do with a tool, not what they announce about it. - [[3. Resources/Literature Notes/The Machine Doesn't Cook for You]] — the sympathetic, identity-threat counter-narrative to the sibling clip's enforcement account of the same non-adoption behavior. - [[3. Resources/Permanent Notes/Work Valued for Its Meaning Resists Automation That Work Valued Only for Its Result Does Not]] — work people do for its meaning, not just its output, resists automation differently than purely instrumental work. - [[3. Resources/Literature Notes/Creativity - How Rest, Wonder, and AI Redefine Productivity with Natalie Nixon]] — rich interview yielding two distinct claims on creative rhythm and burnout. - [[3. Resources/Permanent Notes/Creativity Alternates Wonder and Rigor Rather Than Sustaining Either Alone]] — sustained creative output needs alternating phases of open wonder and disciplined rigor, not either one continuously. - [[3. Resources/Permanent Notes/Work Design Not Tool Speed Decides Whether AI Reduces or Accelerates Burnout]] — the same tool speeds up burnout or relieves it depending on how the surrounding work is designed. - [[3. Resources/Literature Notes/The Work No One Sees (Carl Alexander)]] — cross-linked to the existing fatigue-signal and coding-fatigue-as-enemy-attack notes. - [[3. Resources/Permanent Notes/Masking's Cost Scales Inversely With Its Visibility]] — the more invisible the effort of masking is, the higher its real cost runs unchecked. - [[3. Resources/Permanent Notes/Awareness Without Agency Does Not Produce Sustainability]] — knowing you're burning out doesn't help if you lack the agency to change the conditions causing it. - [[3. Resources/Literature Notes/AI 'Employees' Might Disrupt Work in Unexpected Ways]] — thin, paywalled source; its single claim already exists as the careless-adoption-not-the-technology permanent. ## Synthesis additions (2026-08-16 — labor market and economics) - [[3. Resources/Literature Notes/It's Time to Go Bigger (Theo)]] — central breadth-beats-depth claim already captured by the existing zero-build-cost permanent from the same source; linked, not duplicated. - [[3. Resources/Literature Notes/He Built an AI Executive Assistant in OpenClaw (Atarim)]] — central claim already captured by the existing ops-agent-capability-equals-system-access permanent from the same source. - [[3. Resources/Literature Notes/The One Uncomfortable Truth About AI That Every Educator, Creator, and Business Leader Needs to Hear]] — central claim already captured by the existing template-presence-diagnostic permanent from the same source. - [[3. Resources/Literature Notes/Cursor Quietly Acquires Continue, an Open-Source Alternative to GitHub Copilot]] — permissive open-source licensing preserving code access while a company captures the product value. - [[3. Resources/Permanent Notes/Permissive Licensing Preserves the Code Not the Product]] — a permissive license guarantees you can keep the code; it doesn't guarantee you keep the product built around it. - [[3. Resources/Literature Notes/How to Burst the AI Bubble - Strike at Its Roots]] — Doctorow's labor-law framing linked to the existing bargaining-issue permanent rather than duplicated. - [[3. Resources/Permanent Notes/Judge a Tool by Who Controls It Not What It Does]] — the same technical capability is good or bad for workers depending on who holds the control lever. - [[3. Resources/Literature Notes/Our Vibe Coded Projects That Actually Work (The Vergecast)]] — fuller Lit Note treatment of the source already cited in the 2026-08-11 backlog clearance. - [[3. Resources/Permanent Notes/The Build-Cost Collapse Moves the Bottleneck to Idea Quality]] — once building is nearly free, idea quality — not execution — becomes the binding constraint. - [[3. Resources/Permanent Notes/Personal Fit and Shareability Trade Off in Vibe-Coded Software]] — software built to fit one person's exact workflow tends to be harder to share than software built for many. - [[3. Resources/Literature Notes/If Anyone Can Vibe Code a Product, Who Needs Developers]] — asks what survives once anyone can generate a working product. - [[3. Resources/Permanent Notes/A Moat Is What Survives A Clone]] — the real competitive moat is whatever a cheap clone of your product still can't replicate. - [[3. Resources/Literature Notes/Dr. Jill Lepore on Why AI Backlash Is Vital Decoder]] — long transcript on historical media-democratization cycles applied to the current AI moment. - [[3. Resources/Permanent Notes/New Media Democratizes on Arrival Then Incumbents Recapture It]] — each new medium looks democratizing at first, then incumbents recapture control of it. - [[3. Resources/Literature Notes/Automate, Build, or Buy - The New Agency Playbook]] — largest source in this batch; local-inference observation connected to the existing procurement-spectrum note rather than duplicated. - [[3. Resources/Permanent Notes/Ecosystem Lock-In Insulates Products From Cheap AI Substitution]] — a product's ecosystem lock-in protects it from being undercut by a cheap AI-generated substitute. - [[3. Resources/Literature Notes/Silly Software, Sugar Snap Peas, and Coding in Gen Z Slang w Christin]] — no prior note existed for this source; two new permanents created. - [[3. Resources/Permanent Notes/Cheap Substitutes Raise the Value of the Skill They Replace]] — when a skill becomes trivially substitutable, the judgment to know when the substitute is wrong becomes more valuable, not less. - [[3. Resources/Literature Notes/Very Good Post From the Head of Economics at Anthropic]] — Aaron Levie repost; no new permanent, its claim was already captured by the existing task-automation-does-not-predict-job-elimination note. ## Synthesis additions (2026-08-16 — content and publishing under cheap generation) - [[3. Resources/Literature Notes/You're doing RAG wrong (Akshay)]] — primary claim already existed as a permanent from an earlier pass; Q&A-packet and dedup atoms are mechanisms of that same claim, not new ones. Vendor benchmark figures flagged as unverified. - [[3. Resources/Literature Notes/What if You Were Held Accountable for Your AI Slop]] — cross-referenced into the vault's existing AI-search-integrity cluster. - [[3. Resources/Permanent Notes/AI Output Liability Follows Who Owns the Claim Not Who Disclaims It]] — a disclaimer doesn't move liability off the person who published the claim. - [[3. Resources/Literature Notes/The future of the website]] — Joost de Valk piece yielding two distinct, non-overlapping claims on canonical sourcing and authenticity. - [[3. Resources/Permanent Notes/Canonical Sources Reduce Interface Drift]] — one canonical source rendered through many interfaces drifts less than the same content maintained separately per interface. - [[3. Resources/Permanent Notes/Verifiable Authenticity Becomes a System Property]] — authenticity has to be engineered into the system, not asserted after the fact. - [[3. Resources/Literature Notes/HeyGen AI Tutorial Create Avatar Videos From Just a Script]] — vendor product walkthrough linked to the existing scaled-content-by-production-method note. - [[3. Resources/Permanent Notes/AI Avatar Tools Trade Production Speed for Scrutiny-Proof Quality]] — faster avatar-video production comes at the cost of quality that survives close scrutiny. - [[3. Resources/Literature Notes/This Week in AI - Agents, Gatekeepers, and World Models with Christina Stathopoulos]] — corrects an existing incident note by supplying a missing precondition; the summarization-dependency claim is separately new. - [[3. Resources/Permanent Notes/A Summarizing Intermediary Can Reduce Its Sources to Zero Traffic While Depending on Them]] — an AI summarizer can starve the very sources it depends on for content by intercepting all their traffic. - [[3. Resources/Literature Notes/The Marketing Channel AI Can't Commoditize]] — held in explicit tension against the existing funnel-position note; points the opposite direction (customer-voice vs. frozen-model incumbency). - [[3. Resources/Permanent Notes/AI Recommendation Engines Rank by Customer-Voice Corpus Not Site Authority]] — AI recommendation ranking draws on customer-voice content, not the site-authority signals SEO optimized for. ## Synthesis additions (2026-08-16 — WordPress & CMS disruption) - [[3. Resources/Literature Notes/221 – Rahul Bansal on Using AI Everywhere at rtCamp]] — filename retains the leading "#221" from the source title. - [[3. Resources/Permanent Notes/AI Leverage Requires a Consolidated Data Backbone Built Before the AI Need]] — the data backbone has to exist before the AI need arises; building it reactively is too late. - [[3. Resources/Literature Notes/WP Weekly 295 – Back]] — large ecosystem digest; promoted two of three candidate claims, held back a vendor-sourced bot-traffic stat as unverified. - [[3. Resources/Permanent Notes/A Timed Review Window Trades Update Speed for Supply-Chain Safety]] — a mandatory delay before auto-updating trades faster patching for a supply-chain review window. - [[3. Resources/Permanent Notes/Agent-Readable Data Is Becoming the Default Plugin Integration Surface]] — plugins are increasingly integrating through agent-readable data formats rather than UI automation. - [[3. Resources/Literature Notes/Merge Proposal - Guidelines built on Knowledge]] — two distinct governance claims, each checked against and distinguished from adjacent existing notes. - [[3. Resources/Permanent Notes/Shipping Storage Without Bundled Intelligence Keeps a Platform Vendor-Neutral]] — separating storage from bundled intelligence is an anti-capture move that keeps the platform vendor-neutral. - [[3. Resources/Permanent Notes/A Shared Core Primitive Prevents Per-Plugin Storage Fragmentation]] — one shared storage primitive in core keeps every plugin from reinventing its own incompatible format. - [[3. Resources/Literature Notes/More Targeted Content Editing With Block MCP]] — full deep-read ported; underlies the AI-editing-tools-earn-trust permanent above. - [[3. Resources/Permanent Notes/WordPress's Ecosystem Isolation Forfeits Upstream PHP Fixes]] — WordPress's isolation from the broader PHP ecosystem means it misses upstream fixes the rest of the language already benefits from. - [[3. Resources/Literature Notes/AI-Powered Content and Design, Right Inside the WordPress Editor]] — two distinct claims promoted on platform strategy and layout preservation. - [[3. Resources/Permanent Notes/Shipping Infrastructure Not Features Multiplies Who Can Build]] — shipping reusable infrastructure rather than one-off features multiplies the number of people who can build on top of it. - [[3. Resources/Permanent Notes/AI Edits Should Preserve The Layout You Already Approved]] — an AI content edit shouldn't silently reset layout decisions a human already signed off on. - [[3. Resources/Literature Notes/An introduction to WordPress Studio]] — full appendix already present in the Readwise notes field, ported directly. - [[3. Resources/Permanent Notes/Free Local Tools Funnel Toward Paid Hosting Tiers]] — a free local dev tool's real business purpose is funneling users toward the vendor's paid hosting tier. - [[3. Resources/Literature Notes/WordPress Studio + MCP - A Better AI Workflow]] — colon replaced with " - " in the filename for filesystem safety. - [[3. Resources/Permanent Notes/Agent Tooling Earns Adoption by Augmenting Existing Workflows]] — agent tooling gets adopted when it extends a workflow people already use, not when it demands a new one. - [[3. Resources/Literature Notes/My New Favourite Way to Vibe Code WordPress]] — companion evidence for the existing demo-to-product-gap claim, applied to WordPress specifically. - [[3. Resources/Permanent Notes/A Vibe-Coded Demo Is Easier to Ship Than a Vibe-Coded Product]] — shipping a demo is a much lower bar than shipping the product it demonstrates. - [[3. Resources/Literature Notes/WordPress Studio AI - 5 Powerful Tips]] — full title differs from the shortened filename; recorded in the note's YAML. - [[3. Resources/Permanent Notes/Natural-Language Wrappers Over CLIs Trade Memorization for Verification]] — a natural-language wrapper over a CLI trades the need to memorize commands for the need to verify what it actually ran. - [[3. Resources/Literature Notes/WordPress Studio MCP (Jonathan Bossenger)]] — see the paired permanent under judgment and review, above. - [[3. Resources/Literature Notes/The 5 Levels of Agentic WordPress]] — a maturity ladder for how agentic a WordPress workflow can get. - [[3. Resources/Permanent Notes/Feedback Signals Make Automation Self-Improving]] — automation that receives feedback on its own output improves; automation that doesn't, plateaus. - [[3. Resources/Literature Notes/WP-CLI vs REST API vs Abilities API Choosing the Right WordPress Interface for the Job]] — full deep-read with Refactor Appendix ported directly. - [[3. Resources/Permanent Notes/Caller Distance Determines the Right WordPress Interface]] — how far the caller sits from WordPress core determines which of WP-CLI, REST, or Abilities API is the right interface. - [[3. Resources/Literature Notes/Roadmap 7.1, Gutenberg 23.5, Responsive Styling, Migration to Block Themes — Weekend Edition 368]] — the Core-Abilities claim was linked to an existing permanent; only the migration-pattern claim was new. - [[3. Resources/Permanent Notes/Phased Migrations That Ship a Shared Layer First De-Risk the Rest]] — shipping the shared layer first lets a phased migration de-risk everything that depends on it. - [[3. Resources/Literature Notes/WordCamp US 2026 Publishes Schedule With Expanded AI Programming]] — a distinct source from two other existing WCUS 2026 literature notes already in the vault. - [[3. Resources/Permanent Notes/Flagship Event AI-Track Size Signals Ecosystem Adoption Momentum]] — the size of a flagship event's AI track is a readable proxy for ecosystem adoption momentum. - [[3. Resources/Literature Notes/WordPress Lost Market Share (MemberPress)]] — contrasted with the existing caller-distance note as a different level of analysis (interface selection vs. architecture). - [[3. Resources/Permanent Notes/Callee Interfaces Outlast Caller Integrations Under Provider Churn]] — an interface built on the callee side survives provider churn better than an integration built to a specific caller. - [[3. Resources/Literature Notes/SiteGround WordPress Small-Business AI Report]] — fuller Lit Note treatment of the survey already cited as a raw source in the 2026-08-11 WordPress reference section. - [[3. Resources/Permanent Notes/Adoption Frequency Does Not Measure Output Reliability]] — how often people use AI tells you nothing about whether the output was reliable. - [[3. Resources/Literature Notes/UnblockWP Features]] — vendor marketing page treated as product-claims analysis. - [[3. Resources/Permanent Notes/BYO-Key AI Connectors Keep Governance and Cost With the Site Owner]] — a bring-your-own-key AI connector keeps both governance and cost decisions with the site owner instead of the plugin vendor. ## Synthesis additions (2026-08-16 — maintenance and durability) - [[3. Resources/Literature Notes/Code Maintainability Plummets in the AI Coding Era]] — ported from an existing deep-read. - [[3. Resources/Permanent Notes/AI Coding Optimizes For Prompts, Not Maintainability]] — AI coding output is optimized to satisfy the prompt that produced it, not to stay maintainable afterward. - [[3. Resources/Literature Notes/Maintaining over Building (Brian Coords - Woo)]] — fuller Lit Note treatment of the source already cited as a raw article; one new distinct claim promoted. - [[3. Resources/Permanent Notes/Caution Scales With Install Base Not Timidity]] — being cautious with a widely-installed plugin isn't timidity, it's proportionate to how many sites break if you're wrong. - [[3. Resources/Literature Notes/The AI SDR Playbook]] — vendor content, flagged as such throughout. - [[3. Resources/Permanent Notes/Automation Amplifies a Foundation Rather Than Fixing It]] — automating a sales or ops process amplifies whatever foundation it's built on, good or bad, rather than correcting it. - [[3. Resources/Literature Notes/Can WordPress Studio Code Build This Website (Welcher)]] — evaluates an AI tool against a maintainability-first bar. - [[3. Resources/Permanent Notes/Prefer Core Integration Over Custom Systems for Maintainability]] — building on core-integrated features outlasts building custom systems, even when the custom system is faster to ship. ## Synthesis additions (2026-08-16 — PKM and formation) - [[3. Resources/Literature Notes/Open and Async Work Remote Engineering Culture With Ben Balter]] — both candidate claims promoted; linked to the existing devotional-practice note as a cross-domain parallel rather than a duplicate. - [[3. Resources/Permanent Notes/A Process Whose Value Is the Thinking Cannot Be Automated by Its Output]] — if a process's real value is the thinking it forces, automating its output destroys the value even while preserving the artifact. - [[3. Resources/Permanent Notes/Whether Writing Is Safe Determines Whether an Organization Can Work Openly]] — an organization can only work openly and asynchronously if writing things down doesn't carry personal risk. ## Synthesis additions (2026-08-17, deep-read batch) - [[3. Resources/Literature Notes/Why AI Tools Are Making Us Forget Basic Human Skills]] — AI convenience trades against manual-skill retention by default. - [[3. Resources/Permanent Notes/AI Tool Adoption Removes the Practice That Keeps a Skill Sharp]] — the disuse/atrophy mechanism behind AI-skill-loss claims. - [[3. Resources/Literature Notes/My AGENTS.md & SKILLS.md Breakdown]] — how a skill's always-loaded description field should function. - [[3. Resources/Permanent Notes/Skill Descriptions Are Trigger Keywords Not Documentation]] — a skill's description should read as search-trigger keywords, not explanatory prose. - [[3. Resources/Literature Notes/You Ask, I Answer AI Tools for Messy Notes to Slides]] — staged organize→structure→format prompting for content transformation. - [[3. Resources/Permanent Notes/Staged Prompting Beats One-Shot Prompting for Transformation Tasks]] — splitting a transformation task into stages preserves more source nuance than one combined prompt. - [[3. Resources/Literature Notes/Using Code Review Personas in Claude Code]] — configuring reviewer personas for Claude Code review workflows. - [[3. Resources/Permanent Notes/A Reviewers Point of View Is Context Plus a Stance]] — a reviewer's point of view is mostly context plus a stance, and that's what makes feedback better. ## Synthesis additions (2026-08-18, agent-adjacent development and review) - [[3. Resources/Literature Notes/Democratizing the Theme Directory Building Block-Ready Themes With WP Rig & AI Rob Ruiz]] — WP Rig theme tooling and AI-assisted development; its modular-code/agent-hallucination claim links to the existing Clean Code Is the Human-AI Interface permanent rather than duplicating it. - [[3. Resources/Literature Notes/Designing an Editorial Experience for Humans and AI Alike Aviral Mitaal]] — editorial UX design shared between human editors and AI agents. - [[3. Resources/Permanent Notes/Guardrails Are the Shared Interface Between Human Editors and AI Agents]] — the same UI-level guardrail can serve as the shared interface both human editors and AI agents operate through. - [[3. Resources/Permanent Notes/Semantic Content-Type Naming Reduces Cross-Page Inconsistency]] — naming by content type rather than by page reduces pattern collision and cross-page inconsistency. - [[3. Resources/Literature Notes/Beyond Vibe-Coding Build an AI Pipeline That Ships Production-Ready Plugins Stephane Boisvert]] — a disciplined AI plugin-development pipeline contrasted with unstructured vibe coding. - [[3. Resources/Permanent Notes/A Reviewer That Only Checks Brief-Conformance Penalizes the More Complete Build]] — an evaluation that only checks brief-conformance ranks a more complete, differently-scoped build below a narrower one that matched the brief exactly. - [[3. Resources/Literature Notes/Keynote - The Taste Gap How AI Is Rewiring Developer Experience Bo English-Wiczling]] — WordCamp keynote on AI and developer taste/judgment; all surviving claims connect to the MOC's existing taste-is-the-bottleneck and context-engineering threads rather than duplicating them. - [[3. Resources/Literature Notes/Staff+ Engineers Must Rebuild Team Culture]] — James Socol/LeadDev on AI-enabled parallel work eroding team culture. - [[3. Resources/Permanent Notes/AI Erodes Team Culture by Removing the Friction That Forced Peer Collaboration]] — AI-enabled solo work degrades team culture by removing the everyday friction that used to force the peer interactions culture is transmitted through, distinct from individual-skill-aggregation failure. ## Synthesis additions (2026-08-21) - [[3. Resources/Literature Notes/Writing AI Outreach Emails That Don't Sound Like AI (Benjamin)]] — splits "sounds like AI" into style (fixed via a persistent skill, not a per-conversation prompt) and substance (fixed via Abilities-API-sourced real facts), both subordinate to a human-approval send gate. - [[3. Resources/Literature Notes/Design Systems in the AI Era Competency vs. Craft]] — AI raises design systems' measurable quality floor (consistency, accessibility, performance) without raising the ceiling that makes a product distinctive. - [[3. Resources/Literature Notes/Lost in a Machine Translation Switching From DeepL to OpenAI for a Multi-Brand WordPress Migration]] — swapping a dedicated MT engine for an LLM inside a WordPress multilingual plugin at enterprise scale; what improves and what breaks. - [[3. Resources/Literature Notes/Theo Tests Matt Pocock's Skills Repo and Potato's Pstack]] — tests two third-party AI-coding-agent skill libraries by live use rather than description-reading. - [[3. Resources/Literature Notes/From WordPress Developer to WordPress Engineer Staying Relevant in the AI Era]] — treats accumulated experience as an AI-resistant asset and argues for picking one of four specialty lanes instead of staying a generalist. ## Synthesis additions (2026-08-28 — what a demo omits) - [[3. Resources/Permanent Notes/A Capability Demonstration Is Not an Operating Model]] — the evidence standard separating a working demo from a staffing decision. It concedes the demonstration entirely and rejects only the inference, which is what makes it hard to argue with. The asymmetry matters as much as the logic: the company reverses the experiment, the worker absorbs the discount. - [[3. Resources/Permanent Notes/Coordination Work Bottoms Out on State That Was Never Written Down]] — the complement, and a testable boundary on where assistance stops in cross-team work. Not that coordination is too subtle for a model, but that its inputs — who is available, what was already tried, where the release actually stands — were never written down by anyone. Diagnose the context problem before reaching for a stronger model. - [[3. Resources/Permanent Notes/A Creator Demo That Lets the Product Describe Itself Is Not Independent Evidence]] — registered this run; it was in no MOC. A demo whose conditions the vendor controls carries no information the vendor did not choose to put in it — the evidence-standard sibling of [[3. Resources/Permanent Notes/A Capability Demonstration Is Not an Operating Model]]. - [[3. Resources/Permanent Notes/A Decision Log Is How a Nebulous Role Becomes Inheritable]] — the deliberate counter-move to [[3. Resources/Permanent Notes/Coordination Work Bottoms Out on State That Was Never Written Down]]: publish the decisions and their reasons, accept that the record is partial, and treat the gap between them as the real limit. - [[3. Resources/Permanent Notes/Compressing Access Time Hides Friction Rather Than Removing It]] — the delay was doing two jobs, and only one was overhead. Removing it takes the signal that told you how much you had actually looked at, so a fast fluent answer now needs more scrutiny than a slow one. Corroborates [[3. Resources/Permanent Notes/Sampling a Fraction of a Source Produces False Authority Over the Whole]] with a measured 0.5–5% coverage figure. ## Synthesis additions (2026-10-05 — AI work change and agency pricing) - [[3. Resources/Readwise/Articles/Molly Graham The Grief, Burnout, and Opportunity Hiding Inside the AI Transition]] — On Lenny's Podcast, Molly Graham says delegating to AI is not giving away a Lego because you keep the oversight and the mental tax; the burnout figures (44% in 2025, 55% in 2026) come from the host's own self-selected readership survey. - [AI Is Rewriting the Developer Career Ladder. Here’s How to Stand Out.](3.%20Resources/Readwise/Articles/AI%20Is%20Rewriting%20the%20Developer%20Career%20Ladder.%20Here’s%20How%20to%20Stand%20Out..md) — The GitHub Blog (a vendor selling Copilot) advises developers to direct agents, challenge a model's first answer and spend freed time on bigger problems, with no hiring data behind the career claims. - [[3. Resources/Readwise/Articles/Almost Timely News 🗞️ Highly Opinionated Method of Building AI Agent Skills]] — Christopher Penn (whose firm sells a Skill Fixer) reads his newsletter arguing skills are software, so keep the root SKILL.md short, move deterministic work into scripts, build one-job skills and test activation, outcome and efficiency; his token and determinism figures are unsourced. - [[3. Resources/Readwise/Articles/FlutterFlow - The Missing Layer Between Product and Engineering]] — A LeadDev panel produced with FlutterFlow (two panelists run vendors in this space) says building is cheap so the bottleneck moves to problem definition and code review, with risk-tiered review and a required success metric as controls; the numbers cited are spoken and unsourced. - [[3. Resources/Readwise/Tweets/Prompt of the Day]] — Matt Pocock posts a prompt telling an agent to check his mattpocock/skills v1.3 release, rename CONTEXT.md to GLOSSARY.md, diff the reader's skills against his and scan the last 25 sessions; the release itself is not shown in the capture. - [[3. Resources/Readwise/Articles/AI Is Changing WordPress Agencies Faster Than We Think]] — The WP Minute's show notes for its Agency Action episode summarize two hosts' opinions that AI is shifting agency work from billed hours toward expertise and value-based pricing, with no data in the page. - [[3. Resources/Readwise/Articles/AI Is Changing WordPress Agencies Faster Than We Think-2]] — Two agency owners on WP Minute's Agency Action say a project fell from 20 to 40 hours to 3 to 5 with AI and propose pricing the outcome instead of the hours; the figures are one anecdote and the hosts say the outcome is unknown. - [[3. Resources/Permanent Notes/Hourly Billing Loses Its Basis When AI Compresses Delivery From Days to Hours]] — when AI cuts a deliverable from days to hours, hourly billing shrinks revenue with the work, so agencies have to price the outcome and the risk they carry instead. ## Synthesis additions (2026-10-07 — consumer agent recipes, AI launch coverage and review bottlenecks) - [[3. Resources/Readwise/Articles/School Communication Is Too Much, So My AI Does It]] — Job Thomas describes a scheduled Claude task reading seven school group chats through Beeper's local MCP server and posting a to-do summary to a private chat; he states the never-post-to-groups limit as a prompt instruction, not a tool restriction. - [[3. Resources/Readwise/Articles/Pantheon Launches P1, an AI-Native CMS and Web Operations Platform]] — CMSWire reports Pantheon launched P1 on Oct. 6, 2026, and says, per company statements, that it lets humans and AI agents work on content and code under shared guardrails; features are the company's claims as reported, unverified here. - [[3. Resources/Readwise/Articles/Faster Code Isn’t Faster Delivery]] — Emma Bostian argues in LeadDev that in many teams AI coding tools moved the constraint from writing code to reviewing it, so author time saved becomes reviewer time spent; the evidence is mostly a general-workplace survey, not engineering delivery data. - [[3. Resources/Readwise/Articles/Claude Design How to Create on-Brand Visuals With AI]] — Social Media Examiner's write-up of Jeff Sieh's workflow says a design system built in Claude Design, stored as a design.md file, holds fonts, colors and usage rules, with setup context (open text fields, a website URL) driving output quality; outcome claims rest on Sieh's own production. ## Synthesis additions (2026-10-14 — deep-read batch: agent context and the cost of ownership) - [[3. Resources/Readwise/Articles/Demand-Driven Context What 300 Incidents Reveal About Tribal Knowledge - Raj Navakoti]] — NDC Oslo 2026 talk on scoping an agent's context from its own failures on production incidents; one team's self-reported numbers, no accuracy comparison. - [[3. Resources/Permanent Notes/An Agents Context Should Be Scoped by What It Fails to Know]] — run real tasks, record what the agent says it lacks, and supply only that. - [[3. Resources/Readwise/Articles/Stack Overflow Survey Reactions Software Engineering & Career Insights]] — Overcommitted hosts react to the 2026 survey; the survey figures they cite are spoken-only and unchecked. - [[3. Resources/Permanent Notes/When Code Is Cheap the Cost of Software Is Owning It]] — build-versus-buy turns on maintenance, support, and scaling once code costs little to write. ## Synthesis additions (2026-10-18 — registering previously unlisted notes) ### Permanent Notes - [[3. Resources/Permanent Notes/A Loops Structure Should Be Generated From the Works Structure]] — Theo describes an agent that builds its own loop for the work, merging four stacked PRs overnight, so loop shape should follow the problem. - [[3. Resources/Permanent Notes/A Published Deprecation Schedule Makes a Fast-Moving Dependency Safe to Adopt]] — MCP's 2026-07-28 revision pairs breaking changes with dated removal windows, which the note says makes a fast-moving standard safe to build on. Closest-fit placement. - [[3. Resources/Permanent Notes/A Skill Can Be Evaluated by Pasting Its Text Into a Session]] — To trial a third-party agent skill before installing it, paste its raw markdown into a session and issue its trigger phrase at zero setup cost. - [[3. Resources/Permanent Notes/A Stated Threshold Gates Clarifying Questions Better Than an Open Invitation]] — Telling an agent to ask only when the answer would change the whole output yields fewer, more useful questions than telling it to ask whenever something is unclear. - [[3. Resources/Permanent Notes/A Vendor Retention Policy Change Can Retroactively Break Enterprise Terms]] — A vendor's blanket new data-retention floor can break zero-data-retention and similar enterprise terms customers already negotiated, however clearly the change is announced. - [[3. Resources/Permanent Notes/AI Risk-Aversion Under Uncertainty Rewards Entity Coherence Over Volume]] — When several sources are plausibly correct, AI search defaults to the one with the strongest consistent authority signal, so a focused brand beats a sprawling content library. Closest-fit placement. - [[3. Resources/Permanent Notes/AI-Use Visibility Is a Proxy Screen Not a Competence Signal]] — Andrew Norcross reports a screen that favors visible AI-tool use, which the note argues selects for appearance of trend adoption, not role capability. - [[3. Resources/Permanent Notes/Agent Specification Not Code Is the New Bottleneck]] — Once an agent builder handles the build in minutes, the scarce skill becomes writing a precise specification covering mission, capabilities, data sources, and guardrails. - [[3. Resources/Permanent Notes/Audit Actual Time Before Delegating to AI]] — Auditing how time is actually spent against how it's believed to be spent shows what's worth delegating to AI better than intention does. - [[3. Resources/Permanent Notes/Audit Facts Are Automatable Replaceability Judgment Is Not]] — In a technical-debt audit, a script can gather abandonment, vulnerability, and cost facts, but only a person with business context can judge whether a component is replaceable. - [[3. Resources/Permanent Notes/Casual Self-Hosting Trades a Subscription Fee for an Unvetted Security Liability]] — Replacing a paid SaaS tool with an unvetted open-source deployment converts a known subscription cost into a self-inflicted security risk that AI-assisted building doesn't remove. - [[3. Resources/Permanent Notes/Creative Vision Sets the Ceiling on AI-Assisted Output Quality]] — The quality ceiling of AI-assisted creative work is set by the operator's existing vision, not model capability or prompt detail; without it, output converges on a generic look. - [[3. Resources/Permanent Notes/Curate a Personal Skill Fleet Instead of Bulk-Installing]] — Keep a small, personally reviewed set of agent skills in one synced repository instead of bulk-installing a public library, which works better as a menu. - [[3. Resources/Permanent Notes/Decision Quality Is the Competitive Variable in AI-Augmented Builds]] — When AI handles code generation, a builder's decisions about problem, user, and scope decide outcomes more than build time, shown by a 48-hour hackathon prototype. - [[3. Resources/Permanent Notes/Design-System Maturity Splits Into Measured Competency and Judged Differentiation]] — A design system or AI-assisted creative pipeline matures on two axes: measurable competency that AI can lift, and judged differentiation that needs human taste. - [[3. Resources/Permanent Notes/Government-Imposed Access Restriction Is a Distinct AI Vendor Risk Category]] — A government revoking access to a deployed AI model is a documented risk category that needs evaluating apart from vendor performance, pricing, or deprecation risk. - [[3. Resources/Permanent Notes/In-Band Text Watermarks Are Removable by the Text's Holder]] — A watermark in text characters or token statistics can be stripped by whoever holds the text, so in-band marks fail against motivated holders. - [[3. Resources/Permanent Notes/Multi-Agent AI Projects Need a Persistent Shared Source of Truth]] — Cate ran design, coding, and validation agents as separate sessions, and the note says they need one persistent state file or plan and reality drift. - [[3. Resources/Permanent Notes/Outbound Integration Is Cheap While Inbound Concentrates the Auth Cost]] — Posting into Slack needs only a webhook URL, but receiving replies needs OAuth scopes, event subscriptions, and reinstalls, so integration cost sits on the inbound path. Closest-fit placement. - [[3. Resources/Permanent Notes/Persistent Compute Changes the Developer Boundary]] — Once coding agents outlive a single laptop session, development becomes an operated service that someone has to run, back up, and recover. - [[3. Resources/Permanent Notes/Task Comprehension Bounds Delegation Quality]] — You can't delegate work you can't specify, so prompting an AI well and managing a person well are the same skill applied to different delegates. ### Literature Notes - [[3. Resources/Literature Notes/AI Influence Level (AIL) V1.0]] — Daniel Miessler's AI Influence Level is a compact ordinal label, AIL0 to AIL5, disclosing how much AI shaped a text without judging its quality. - [[3. Resources/Literature Notes/AI Models Can Disappear Overnight. Is Your Engineering Team Built to Survive It]] — Argues teams should check how tightly AI features couple to one provider, since models can vanish overnight, and reports one data point on AI logic share. - [[3. Resources/Literature Notes/Alliance for America's Skilled Trades (Fortune Op-Ed)]] — A Fortune op-ed announces an alliance of companies including Ford, Carhartt, BlackRock/GIP, and Google to promote skilled trades. Closest-fit placement. - [[3. Resources/Literature Notes/Almost Timely News - How To Do Feature Engineering with AI (2026-07-12)]] — Christopher Penn argues survey design should ask only for fields that can't be inferred, then hand enrichment-pipeline building to an AI coding agent under a PII rule. - [[3. Resources/Literature Notes/Asking Cursor to Review My Blog for Performance]] — Raymond Camden asked Cursor to review a blog for performance; one finding was a dead Mastodon-shortcode endpoint holding builds open about 70 seconds without a timeout. - [[3. Resources/Literature Notes/Build Your Own YouTube Competitor Dashboard for FREE]] — A creator gives away a master prompt and ten reference screenshots that build a YouTube competitor dashboard in Lovable, as a vibe-coding case study. - [[3. Resources/Literature Notes/Claude Just Dropped a Design Tool That Prints Money (Terry Brock)]] — Terry Brock's video walks through Claude Design, which turns prompts into slides, prototypes, and documents; the note treats the money-making framing as unverified vendor-demo hype. - [[3. Resources/Literature Notes/Developers on the frontline of the SaaS replacement wave]] — LeadDev reports a real but uneven wave of SaaS replacement, where the hard parts have moved to data extraction and the security cost of self-hosting. - [[3. Resources/Literature Notes/Don't Use AI You Might Not Get the Job]] — Andrew Norcross says a job was lost over too little AI use, and the note argues visible AI adoption is a weak hiring signal beside verification skill. - [[3. Resources/Literature Notes/GitHub Copilot Chat in Visual Studio 2022]] — GitHub Copilot Chat in Visual Studio 2022 adds conversational help beyond completion, using surrounding code context for explanations, analysis, and some fixes. Closest-fit placement. - [[3. Resources/Literature Notes/He Built 10up to $45M-yr; AI Brings Back Web Dev's Golden Era (Jake Goldman)]] — Jake Goldman, who built 10up, argues AI brings back a golden era of web development for curious generalists, and the note raises liability questions. - [[3. Resources/Literature Notes/How Anthropic Uses Claude Code - Agentic Software Engineering at Scale]] — Daisy Hollman argues context engineering now binds agentic engineering more than model capability, so each tool or skill should cost nothing when idle. - [[3. Resources/Literature Notes/How Jupyter AI Brings Agentic Workflows Into Notebooks]] — Lior Turetsky says Jupyter AI v3.0 adds notebook-native editing tools, real-time collaboration, and open protocols so notebooks work with agents across providers. - [[3. Resources/Literature Notes/How To Add Skills to Claude AI - Text Humanizer]] — A short tutorial shows how to install a third-party Agent Skill in Claude's web app by downloading a zip from GitHub and uploading it in settings. - [[3. Resources/Literature Notes/How the Anthropic Saga Could Threaten American AI Dominance]] — Argues export controls on Anthropic models show political vendor lock-out as an AI infrastructure risk, with open-weight fallback and multi-provider abstraction as responses. - [[3. Resources/Literature Notes/How to Connect Slack to Your Bolt.new App]] — A Bolt.new tutorial argues Slack apps make a vibe-coded project two-way: webhooks push events out, buttons and threads react, and results reflect back in real time. - [[3. Resources/Literature Notes/How to Demonstrate Your AI Chops in Job Hunting (Almost Timely News)]] — Christopher Penn shows how to demonstrate AI skills when job hunting, and the note questions the evidence, which rests on one demonstration. - [[3. Resources/Literature Notes/How to stay technical as an engineering manager]] — Emma Bostian says engineering managers must stay technical and lists four strategies: structured AI prompting, sparing pairing, one deep domain at a time, and architecture review. - [[3. Resources/Literature Notes/I guess we're writing loops now (Theo)]] — Theo reports that agentic coding loops work within bounds: shape them from the work, review agent-to-agent first, and accept that token cost limits who can run them. - [[3. Resources/Literature Notes/Keynote - After the AI Hype, What's Real and What's Next (Richard Campbell)]] — Richard Campbell's keynote separates real AI value from hype, contrasting AlphaFold with chatbots and noting a gap between announced and built data-center capacity. - [[3. Resources/Literature Notes/Lit · AI interview-practice platforms create false confidence]] — Claims AI interview-practice platforms give false confidence because chatbots are sycophantic, and suggests adversarial personas to build transferable confidence. Closest-fit placement. - [[3. Resources/Literature Notes/Lit · AI tools democratize WordPress block development]] — Telex and similar AI tools let non-coders build niche WordPress blocks, so coding skill stops being the differentiator and taste matters more. - [[3. Resources/Literature Notes/Loop Engineering (The New Stack) — Boris Cherny · Addy Osmani]] — Boris Cherny and Addy Osmani describe loop engineering with six primitives: automations, worktrees, skills, connectors, sub-agents, and memory, plus comprehension debt. - [[3. Resources/Literature Notes/Mosh, Caddy & Tailscale - A Remote Dev Deep Dive]] — The Syntax hosts say long-running coding agents push development onto an always-on host reached by Mosh, Caddy, and Tailscale, which then needs operating. - [[3. Resources/Literature Notes/Navigating Claude Code - The Full Workflow]] — A ten-article series walks through the Claude Code workflow, with command names and cost figures tied to a 2026 release and flagged for verification. - [[3. Resources/Literature Notes/Navigating Claude Code Agent Teams in Practice]] — Agent teams earn their coordination cost only for parallel, loosely coupled work like competing hypotheses; routine tasks do better with a single subagent. - [[3. Resources/Literature Notes/Navigating Claude Code Subagents Done Right]] — Oleg Efimov says subagents save main-session context only when the task is scoped, tools and model are restricted, and the report comes back short and structured. - [[3. Resources/Literature Notes/OpenClaw VPS Guide - How to Deploy OpenClaw On xCloud]] — A video walkthrough shows deploying the beta OpenClaw assistant on xCloud in about five to seven minutes at $28.80 a month, with ChatGPT OAuth and Telegram. Closest-fit placement. - [[3. Resources/Literature Notes/Pluralistic - AI and amateurism (Cory Doctorow, 15 Jun 2026)]] — Cory Doctorow argues vibe coding suits personal tools people build for themselves but turns dangerous when unreviewed code goes into production or commercial software. - [[3. Resources/Literature Notes/Syntax Weekly — Anthropic Doesn't Use AI, HTML Streaming in Chrome]] — A weekly roundup covering Anthropic's no-AI promo video, an export ban discussion, and HTML streaming in Chrome, with the ban's trigger flagged as unverified. Closest-fit placement. - [[3. Resources/Literature Notes/Technical debt is becoming an AI problem]] — Jono Alderson argues technical debt is the cost of future change, so organizations that reduced complexity adapt to any AI standard more cheaply than those betting on one. - [[3. Resources/Literature Notes/The AI-augmented leader - Architecting your own virtual chief of staff]] — A talk argues leaders who audit how they actually spend time find the gap shows what to delegate to AI as a virtual chief of staff. - [[3. Resources/Literature Notes/The weird situation with Fable]] — A creator argues Anthropic's Fable release billed full price for invisibly degraded output and edited system-card documentation afterward without a changelog. - [[3. Resources/Literature Notes/Vibe coding can build your pipeline. It can't explain it six months later]] — Argues vibe coding can build a pipeline but can't explain it later, so versioned specs should be the system; the author sells spec tooling. - [[3. Resources/Literature Notes/We're Announcing the Alliance for America's Skilled Trades (Google)]] — Google, BlackRock, Carhartt, and Ford launched the Alliance for America's Skilled Trades on July 21, 2026, aggregating existing workforce training commitments across 30 states. Closest-fit placement. - [[3. Resources/Literature Notes/WordPress + Claude — a new look for cate.blog]] — AI design and coding tools speed up a redesign, but WordPress's database-backed configuration and split CSS ownership absorb most of the gain. - [[3. Resources/Literature Notes/Zero to Agent in 30 Minutes - Build a YouTube Analytics Agent with Vicki Reyzelman]] — A talk builds a YouTube analytics agent in 30 minutes, with a skills file listing mission, capabilities, data sources, guardrails, and acceptance criteria. ## Synthesis additions (2026-10-19 — GameDev Day talks and agent-selection reading) - [[3. Resources/Readwise/Articles/Fix Your CI Without Leaving the IDE TeamCity MCP and CLI - Artem Rokhin - GameDev Day 2026]] — Artem Rokhin (TeamCity) argues that one interface, the IDE or terminal, lets a developer resolve CI problems without switching to a browser. Covers the TeamCity CLI, an MCP server with scoped tokens, and configuration stored as code. The productivity claim is untested in the source. - [[3. Resources/Readwise/Articles/Agentic Development in Unreal Engine With JetBrains Rider - Alexander Dzerakh - GameDev Day 2026]] — Alexander Dzerakh presents Rider Kit, an AI chat inside the Unreal Editor that connects editor state to the Rider IDE so a coding agent sees both. A product preview: the performance claims aren't measured. - [[3. Resources/Readwise/Articles/Live-Building AI-Driven PCG With Godot, C, and Gemini - Sharmishtha Sen - GameDev Day 2026]] — Sharmishtha Sen connects a local Godot MCP server to cloud and local models so assistants in Rider write compilable Godot 4 C# for procedural terrain. Slide narration, not live coding; no repo or benchmarks. - [[3. Resources/Readwise/Articles/How We Automated Figma-to-Unity UI Integration With AI Agents - GameDev Day 2026]] — Alina and Olia Chupakhina describe a Figma-to-Unity pipeline: an MCP server, six agent skills and a deterministic converter. They report about 4x faster UI work; the comparison table is on a slide the transcript doesn't carry. - [[3. Resources/Readwise/Articles/How AI Impacts Podcasting - Watch Me Build a Podcast App]] — The Podcast Setup Newsletter narrator builds a native podcast app with an AI coding agent and reports a build that took days in 2025 finishing in about 10 minutes. The 10-minute run cloned the finished 2025 site, so the comparison is confounded. - [[3. Resources/Readwise/Articles/This Is Why WordPress Belongs on an AI Agent’s Website Shortlist]] — Remkus de Vries reports an agent benchmark in which zero of 193 completed builds used WordPress, and argues WordPress needs to be a legible choice for agents. The author runs WordPress businesses; the benchmark data wasn't checked. ## Synthesis additions (2026-10-19 — Permanents from the GameDev Day, agent-site and religious-history batch) - [[3. Resources/Permanent Notes/Applying Only the Diff to a Generated Artifact Keeps Scripts and Hand Fixes a Rebuild Erases]] — When a generated artifact has gained hand-written scripts, cross-references and manual fixes, an update that compares it with the changed source and applies only the differences keeps them. ## Synthesis additions (2026-10-20 — deep-read batch) - [[3. Resources/Readwise/Articles/Harness - What Breaks When AI Opens Pull Requests]] — A LeadDev panel made in partnership with Harness. Larry Garfield says AI doesn't change GitOps because a pull request gets the same tests and review whoever wrote it; Martin Reynolds (Harness field CTO) says AI helps at several pipeline stages. The title promises more about AI-opened pull requests than the GitOps discussion delivers; the statistics are unsourced.