Agents & The New Internet (35) - CourtneyR.dev's 2nd Brain

Agents & The New Internet (3/5) 🔗 **Source:** [Agents & The New Internet (3/5)](https://j.cv/agents-the-new-internet-3-5/) > [!note]+ Document Notes > tags: [AI, web, agents] > author: [J.CV] > source: [j.cv] > reference: [https://j.cv/agents-the-new-internet-3-5/] > date created: 2026-01-27 12:00 > > # 🏷️ Title: Agents & The New Internet (3/5) >  ## 🧭 Table of Contents - [📝 Summary](#-summary) - [📖 Main Content](#-main-content) - [🗒️ Notes](#️-notes) - [📚 References](#-references) - [🗣️ Paraphrase](#️-paraphrase) - [📰 Newsletter](#-newsletter) - [✅ Action Items](#-action-items) - [🧪 Hypotheses](#-hypotheses) - [📊 Methodology](#-methodology) - [❓ Questions](#-questions) - [🔮 Future Research](#-future-research) - [💡 Implications](#-implications) - [🧠 Learning Styles Assessment](#-learning-styles-assessment) - [📊 Accessibility Assessment](#-accessibility-assessment) - [🌍 DEIB Assessment](#-deib-assessment) - [✅ Logical Validity](#-logical-validity) - [🛡️ DARVO Analysis](#-darvo-analysis) - [🎯 Conclusion](#-conclusion) ### 🔗 References - Include links cited in or relevant to the article. ## 📝 Summary - The article argues the web is shifting from AI as a Q&A intermediary to autonomous agents that pursue goals, traverse sites, and take actions on users’ behalf. - This matters because websites become machine-facing data and capability surfaces, not just human-facing pages. - Stakeholders: publishers, SaaS/product companies, commerce, media, infra providers, browser/AI vendors. - Risks: loss of control over representation, commoditization of content, uneven readiness for agent consumption. - Bottom line: structure content and expose capabilities now so agents can understand, trust, and act with/for you. ## 📖 Main Content - The author contrasts “intermediary” AI (Perplexity-style synthesis) with “agentic” AI where users delegate goals and agents handle discovery, comparison, and actions like booking or purchasing ([From Questions to Goals](https://read.readwise.io/read/01kfy0trartxxaw2zaaynny4pz?endLexposition=0000I_00000-004B&startLexposition=0000A&emphasizeText=The+agent+does+the+traversal%2C+the+synthesis%2C+the+comparison%2C+and+potentially+the+action+itself)). - Agents still rely on today’s web stack: search indices, domain authority, links, and site content; what changes is that agents, not humans, increasingly do the traversing and deciding ([How Agents Will Traverse the Web](https://read.readwise.io/read/01kfy0trartxxaw2zaaynny4pz?endLexposition=0000S_00000-004f&startLexposition=0000K&emphasizeText=Agents+will+use+the+same+infrastructure+the+web+already+has)). - Three interaction modes emerge: browser automation, programmatic access (e.g., MCP-style tools/APIs), and generative browsers that transform multiple sites into custom interfaces (e.g., Google Disco GenTabs, Perplexity Comet, OpenAI Atlas) ([Google’s Disco demonstrates this shift](https://read.readwise.io/read/01kfy0trartxxaw2zaaynny4pz?endLexposition=0000e_00000-006h&startLexposition=0000U&emphasizeText=Google%E2%80%99s+Disco+demonstrates+this+shift)). - For publishers, this means their “pages” become raw material for generated comparison tools and apps; discovery and presentation layers are rebuilt on the fly, outside their control ([The implications for publishers are significant](https://read.readwise.io/read/01kfy0trartxxaw2zaaynny4pz?endLexposition=0000i_00000-002P&startLexposition=0000g&emphasizeText=Your+content+becomes+raw+material+for+interfaces+you+didn%27t+design+and+don%27t+control)). - Agency evolves from today’s supervised synthesis to bounded-task execution and eventually more autonomous agents governed by standing guidelines; humans stay “in the loop” via policy and review, not step-by-step approvals ([The Evolution of Agency](https://read.readwise.io/read/01kfy0trartxxaw2zaaynny4pz?endLexposition=0000o_00000-003F&startLexposition=0000k&emphasizeText=Further+out%2C+agents+operate+more+autonomously+based+on+standing+guidelines)). - Sites must now serve both humans (with richer, AI-powered experiences) and machines (with structured, semantically dense, well-linked content and clear action surfaces) ([What This Means for the Web](https://read.readwise.io/read/01kfy0trartxxaw2zaaynny4pz?endLexposition=00014_00000-004W&startLexposition=0000q&emphasizeText=If+agents+are+doing+the+discovery%2C+comparison%2C+and+action)). - The “representation problem”: agents will summarize and reframe your offerings according to their own logic; for commodity commerce this may be fine, but media and services risk losing brand, voice, and nuance ([The Representation Problem](https://read.readwise.io/read/01kfy0trartxxaw2zaaynny4pz?endLexposition=0001I_00000-003Z&startLexposition=00016&emphasizeText=The+agent+becomes+the+interface+between+you+and+the+user)). - Near term, the practical move is to structure static content for agent consumption: clean semantic markup, clear hierarchies, progressive disclosure, and explicit metadata about authority and purpose ([The Next Year](https://read.readwise.io/read/01kfy0trartxxaw2zaaynny4pz?endLexposition=0001W_00000-006D&startLexposition=0001K&emphasizeText=Structure+your+content+so+that+when+an+agent+arrives)). - Longer term, sites will deploy their own “delegates” – agents that understand their content, constraints, and goals, and can converse and negotiate with visiting user agents using protocols like MCP and Agent2Agent ([The Need for Representation](https://read.readwise.io/read/01kfy0trartxxaw2zaaynny4pz?endLexposition=0001o_00000-0024&startLexposition=0001a&emphasizeText=if+agents+are+going+to+represent+you+to+users)). ### 🗒️ Notes - AI on the web is evolving from answering user questions to executing user goals; this shifts responsibility for synthesis and action from humans to agents. - Intermediary model ≈ “smart search”: user asks, AI reads sites, returns a synthesized answer; user still decides and acts. - Agentic model: user states a goal (e.g., “book flight,” “hire plumber”), agent discovers options, compares, and may execute transactions, returning only confirmation. - Agents rely on existing search engines (Bing, Brave, Google) plus domain authority and link structures; they sit atop, not outside, today’s web infra. - Interaction modes: - Browser automation: agents “drive” a headless browser, clicking and filling forms—universal but slow and brittle. - Programmatic: sites expose capabilities via protocols like MCP/tools APIs, letting agents call actions directly with structured IO. - Generative browsers: tools like Google Disco GenTabs, Perplexity Comet, OpenAI Atlas consume multiple open tabs and generate a bespoke interface or app. - Generative browsing breaks the page-centric discovery model: sites become data sources for synthesized, task-specific UIs that users may treat as the primary experience. - Publisher implications: - Carefully designed product or article pages may never be viewed; instead, their content fuels comparison tables, planners, or dashboards. - “Discovery layer” is reconstructed dynamically by agents, not SEO or feed-based UIs alone. - Agency trajectory: - Today: supervised synthesis and suggestion. - Next: explicit task delegation with bounded authority (e.g., purchases under a limit, bookings within constraints). - Later: semi-autonomous agents that act under general policies and report outcomes periodically. - Human role evolves from micromanagement to policy-setting and results review; “approval” moves from each click to governance and monitoring. - Websites must dual-serve: - Humans: richer, more personalized experiences, often AI-augmented, as mere text pages become undifferentiated. - Machines: structured content (semantic HTML/markdown), clear schemas, internal links, long-tail SEO-like practices generalized for LLMs. - AIO/GEO “optimization” is, in the author’s view, largely rebranded long-tail search optimization focused on making content ingestible and semantically rich. - Agents that take actions need: - A clear map of possible actions (capabilities). - Programmatic interfaces (APIs, tools, MCP servers) to invoke them. - Trust signals (reputation, security, provenance) to choose which entity to interact with. - Representation problem: - Human visitors see your designed experience; visiting agents see only extracted facts and metadata. - Agents then present you to users in blended summaries where brand, voice, and differentiation can be flattened. - Commerce vs media: - Commerce: if the description is accurate and the sale occurs, lack of site “visit” may be acceptable. - Media/services: a flattened summary can erase editorial stance, nuance, and identity—core value props. - Short-term, agent-readiness: - Use structured markdown and accessible semantic HTML. - Organize pages like briefings: start with key points, include clear sections, show relationships between pieces of information. - Rank information by authority and importance; favor progressive disclosure (summary → detail). - Add metadata (e.g., author, date, canonical, type, confidence) that helps agents assess authority and relevance. - Long-term “full delegate” vision: - Your site runs its own agent that understands offerings, pricing, constraints, SLAs, and policies. - Visiting user agents negotiate and query this delegate instead of scraping pages. - Underlying infra is emerging: MCP (now under the Linux Foundation with major vendors) and Agent2Agent protocols for agent-to-agent communication. - Strategic message: entities that structure content, expose capabilities, and plan representation in an agent-to-agent world will gain advantage; laggards will remain scrapeable but voiceless. ## 📚 References - J.CV – “Agents & The New Internet (3/5)” (source article) - Model Context Protocol (MCP) – Linux Foundation AI & Data - Perplexity AI – Perplexity Search & Comet - Anthropic – Claude, Computer Use / Cowork - OpenAI – ChatGPT browsing, Atlas browser experiments - Google – Google Search, Google AI, Google Disco / GenTabs - Brave – Brave Search - Early Agent-to-Agent protocol discussions (various OSS/standards repos) ## 🗣️ Paraphrase - The web is moving from AI that simply answers questions to AI agents that can pursue user goals end-to-end, using existing web infrastructure to search, compare, and act on our behalf. - As agents become the primary visitors, websites must shift from being human-optimized pages to machine-readable data sources and capability surfaces, and eventually to active agents that can represent their owners in agent-to-agent negotiations. ## 📰 Newsletter AI on the open web is quietly rewiring what a “site visit” even means. In part 3 of this series, J.CV explores how agents move from answering questions to executing goals—using your content as raw material for custom interfaces and automated decisions. For WordPress builders, this raises a new frontier: making your sites not only beautiful for humans, but legible and actionable for agents. That means: - Clean semantic markup and structured content instead of ad‑hoc layouts. - Clear action surfaces via APIs, tools, and well-documented workflows. - Rich, AI-assisted experiences for human visitors that go beyond static pages. If you’re working on plugins or themes, consider how your data models, REST endpoints, and block structures will look to an autonomous agent tasked with “compare three options and choose the best.” The time to design for that audience is now. ## ✅ Action Items #### DevRel - Map your product’s core workflows (“book,” “buy,” “configure,” “compare”) to explicit, well-documented APIs or tools that agents can call. - Publish examples and quickstarts that show how to integrate your services into agent toolchains (e.g., MCP servers, SDK helpers). - Run workshops on “designing machine-legible docs” focused on semantic structure, schemas, and progressive disclosure for agents. - Track adoption of your APIs/tools in popular AI platforms; set KPIs for agent-originated traffic or transactions. - Collect developer feedback on friction agents hit (auth, rate limits, unclear responses) and feed into DX improvements. #### OSPO - Define policies for exposing internal capabilities to external agents (auth, data boundaries, rate controls). - Ensure open-source components used for agent integration (MCP servers, SDKs) have clear licenses and compliance checks. - Establish a vulnerability intake and embargo policy specifically for agent-facing endpoints. - Maintain an inventory of which internal systems are callable by agents, with ownership and freshness requirements. - Engage with upstream projects (MCP, Agent2Agent, major AI SDKs) to influence standards that affect your interoperability. #### Community - Offer maintainers guidance and templates for making projects agent-friendly (OpenAPI specs, MCP tool definitions). - Create contributor paths around “AI readiness” – improving docs, schemas, and APIs for agent consumption. - Share best practices with sponsors and partners so their integrations remain effective as agent usage grows. - Host cross-project discussions on standardizing trust and representation signals for agents. #### WordPress - **Core:** explore first-class patterns for semantic content and structured data that benefit both search and agents. - **Core Performance:** measure and optimize the cost of agent-driven traffic (headless browsing, API hits). - **Core AI:** define guardrails and APIs that let agents interact safely with WordPress data and actions (CRUD, publishing, commerce). - **Core Program Team:** coordinate experiments around “agent-ready” themes/plugins and document learnings. - **Mobile:** ensure app APIs expose key user actions in a way agents can safely orchestrate. - **Playground:** provide agent-focused sandboxes to test how autonomous tools interact with WordPress sites. - **Community:** add “AI/agent readiness” tracks to meetups and WordCamps. - **Training:** develop curricula on building machine-legible WP content and APIs. - **Marketing:** frame WordPress as a strong base for agent-compatible websites and headless experiences. - **Support:** gather tickets related to bots/agents and feed patterns back into design and security. - **Design:** create patterns for content layouts that remain clear in both human and machine consumption paths. - **Documentation:** expand DevHub/HelpHub with examples of exposing capabilities to agents (REST, WP-CLI, MCP-like tooling). - **Photos / TV / Openverse:** ensure media metadata is rich and consistent so agents can accurately select and attribute assets. - **Polyglots:** consider how multilingual content and glossaries appear to agents; maintain consistent semantics across locales. - **Accessibility:** ensure AI-enhanced experiences remain accessible as interfaces become more personalized/dynamic. - **Meta / CLI / Hosting / Tide / Themes / Plugins:** align scoring, guidelines, and tooling with “agent-ready” practices (structured data, trustworthy APIs, clear provenance). #### The WPOCC - Design funding and fellowship programs around upgrading widely-used projects and sites for agentic web readiness. - Sponsor documentation sprints to create machine-legible patterns and governance templates for AI integration. - Pilot transparent funding models for shared infra like MCP servers or Agent2Agent gateways in the WP ecosystem. ### FAIR - Encourage technically independent verification of agent actions via signed metadata and auditable logs. - Promote transparent funding/disclosure for AI integrations to avoid capture by any single vendor. - Turn governance patterns (acceptable use, delegation limits, review processes) into reusable templates for AI deployments. ## 🧪 Hypotheses - H1: Websites with well-structured, semantically rich content will be favored by autonomous agents, leading to higher agent-driven conversions. - H2: Exposing programmatic capabilities (e.g., MCP tools, APIs) will increasingly matter more than traditional UI polish for commercial outcomes. - H3: Media organizations that fail to design for agent representation will see brand differentiation erode in AI-generated summaries. - H4: Agent-to-agent protocols will create a new interoperability layer analogous to, but distinct from, traditional HTTP+HTML. ### 📊 Methodology - For H1: Compare traffic and conversion metrics from known agent referrers between structured vs unstructured sites in similar niches. - For H2: A/B test sites with UI-only vs UI+public APIs for agents, tracking bookings/sales attributed to automated clients. - For H3: Content analysis of AI summaries across outlets with varying levels of structured data; measure preservation of brand-specific framing. - For H4: Track adoption and usage patterns of MCP/Agent2Agent implementations across vendors; analyze emergent standards and ecosystem clustering. ### ❓ Questions - How will agents arbitrate between conflicting representations of similar products or claims? - What trust and reputation systems will agents rely on beyond current domain authority? - How can entities express “terms of interaction” to agents (e.g., negotiation ranges, risk limits) in machine-readable form? - What legal obligations arise when agents misrepresent entities or users in negotiations? ### 🔮 Future Research - Standard schemas for agent-readable representations of capabilities, constraints, and preferences. - Governance models for user-agent and site-agent relationships, including revocation, oversight, and liability. - UX patterns for humans supervising semi-autonomous agents that operate across multiple sites and services. - Economic analysis of how agent-mediated commerce changes platform power dynamics. ### 💡 Implications - Practice: Product and content teams must design simultaneously for human UX and machine ingestion. - Policy: Regulators may need to address agent transparency, misrepresentation, and negotiation fairness. - Security: Agent-facing endpoints and tool APIs expand attack surfaces and require dedicated hardening. - Education: Developers and marketers need training in “agent-ready” information architecture and protocol design. - Funding: Infrastructure for protocols like MCP and Agent2Agent becomes a strategic investment area. ### 🧠 Learning Styles Assessment - Reading level estimate: roughly upper high-school to early undergrad; conceptually advanced but written in plain, jargon-light prose. #### 👀 Visual learners - Present: conceptual contrasts (intermediary vs agentic), examples of tools and flows. - Improve: diagrams of agent workflows, browser vs programmatic vs generative modes, and agent-to-agent negotiations. #### 🔊 Aural learners - Present: none inherent. - Improve: short audio summary or podcast segment walking through concrete user scenarios. #### 🗣️ Verbal learners - Present: clear definitions and analogies (e.g., “pile of documents vs well-organized briefing”). - Improve: a brief glossary (agents, delegates, MCP, agent-to-agent) with side-by-side comparisons. #### 🧩 Kinesthetic learners - Present: mostly conceptual discussion. - Improve: hands-on labs building a simple MCP tool, instrumenting a site for agent access, or testing with an existing browser agent. #### 🧠 Logical learners - Present: stepwise progression of agency, causal reasoning about infra and incentives. - Improve: explicit flowcharts and decision trees for how agents choose sites or actions. #### 🧑‍🤝‍🧑 Social learners - Present: implications for organizations and ecosystems. - Improve: guided discussion prompts or case studies for teams designing agent strategies. #### 🧘‍♀️ Solitary learners - Present: reflective, forward-looking essay style. - Improve: self-check questions at the end of sections (“How would my site look to an agent today?”). #### 🌿 Naturalistic learners - Present: no explicit environment/sustainability links. - Improve: analogies to ecosystems (agents as species, protocols as habitats) and discussion of compute/energy impacts. ### 📊 Accessibility Assessment *(Assuming a typical modern web article implementation)* #### Screen reader semantics - Present: headings and paragraphs. - Improve: ensure consistent heading hierarchy, landmark regions, and descriptive link text. #### Keyboard navigation - Present: basic scroll and link navigation. - Improve: verify focus outlines, skip links, and absence of keyboard traps in any interactive enhancements. #### Contrast and readability - Present: likely adequate for text content. - Improve: check contrast ratios for body and accent text; ensure generous line height and spacing. #### Text resizing and reflow - Present: responsive layout. - Improve: verify 200% zoom without loss of content or functionality; support reduced-motion preferences. #### Media and captions - Present: mostly text; a few referenced tools. - Improve: if adding visual diagrams or videos, provide alt text, captions, and transcripts. #### Forms and interactions - Present: minimal or none. - Improve: if adding interactive demos, ensure labeled controls and accessible error messages. #### Standards reference - Map any gaps against WCAG 2.2 AA, especially 1.4.x (contrast), 2.1.x (keyboard), and 1.3.x (structure). ### 🌍 DEIB Assessment #### Cultural breadth - Present: primarily Western/US-centric tech ecosystem examples. - Improve: highlight global implications and non-Western platforms or regulatory contexts. #### Inclusive language - Present: generally neutral, technical language. - Improve: add brief plain-language summaries and minimize niche jargon where possible. #### Neurodiversity support - Present: clear structure and chunked sections. - Improve: more consistent subheadings, callout boxes for key points, and minimized long paragraphs. #### Representation - Present: abstract discussion; no people imagery. - Improve: when adding visuals, ensure diverse representation and inclusive alt text. #### Psychological safety - Present: non-confrontational, speculative tone. - Improve: community guidelines and norms if this essay anchors discussion spaces. ### ✅ Logical Validity #### Strong reasoning - The progression from intermediary to agentic AI is argued causally: as capabilities extend from reading to acting, delegation levels naturally increase. - The claim that structured content helps agents is grounded in analogy to SEO and search indexing, a well-understood precedent. - The representation problem follows from control shifting from publisher-curated UIs to agent-curated summaries; the loss of control is logically consistent. #### Fallacies - Some extrapolations (e.g., ubiquity of site delegates) are speculative but framed as plausible futures, not inevitabilities; no major fallacies stand out. - Risk: mild tech determinism—assumes adoption trajectories without deeply engaging with regulatory or economic brakes; this could be balanced with counter-scenarios. #### Tone of voice - Analytical, forward-looking, and pragmatic. - Uses concrete examples (trip booking, plumbers, investment research) to ground abstract claims. ### 🛡️ DARVO Analysis - Deny: No denial or minimization of harms; the author acknowledges real risks (loss of representation, flattening of brands). - Attack: No attacks on critics or alternative views; other approaches (SEO, AIO/GEO) are described critically but not ad hominem. - Reverse victim/offender: No attempt to portray powerful actors as victims; instead, the piece focuses on how entities can adapt to structural changes. ## 🎯 Conclusion The article sketches a near-future web where agents, not humans, are the primary visitors and decision-makers. To stay relevant, organizations must restructure content, expose capabilities programmatically, and eventually deploy their own delegates so they’re represented fairly in agent-to-agent interactions. Practical next steps: audit how machine-readable your site is today, prioritize structured content and clear APIs, and begin defining how you want agents to understand and act on your behalf. --- ## 🧱 Refactor Appendix (append after 🎯 Conclusion) ### 📚 Literature Split Pad - Purpose: Stage Konik CEE units that become per-source Literature notes. - Split level: **H3**. - Include verbatim quotes under 25 words and a [view highlight] link. ### [H1] 5–10-word present-tense claim > [!quote] [{Title or entity}](https://j.cv/agents-the-new-internet-3-5/) by [[{Author or [MISSING]}]] via [[{Site or [MISSING]}]]. Published {DATE_OR_[MISSING]}. Accessed . > “Short quote, verbatim, ≤25 words.” Two lines: context and use. Cite data if relevant. [view highlight](https://j.cv/agents-the-new-internet-3-5/#H1) #xref #articleseed ID:rw_[MISSING] ### [H2] 5–10-word present-tense claim > [!quote] [{Title or entity}](https://j.cv/agents-the-new-internet-3-5/) > “Short quote, verbatim, ≤25 words.” Two lines: context and use. [view highlight](https://j.cv/agents-the-new-internet-3-5/#H2) #xref ID:rw_[MISSING] --- ### 🔖 Atomic Split Pad (Zettel Candidates) - Purpose: Promote reusable claims into Atomic notes. - Split level: **H4**. - Each item: one idea, present tense, specific, testable. #### [Z1] 5–10-word present-tense claim - Scope/condition in one line. - Pointer to page/time or dataset. - Links: [[Parent-Concept]] [[Related-Idea]] #### [Z2] 5–10-word present-tense claim - Scope/condition in one line. - Pointer to page/time or dataset. - Links: [[Parent-Concept]] [[Related-Idea]] --- ### ✏️ Fleeting Queue (Extract by Selection) - Purpose: Capture quick actions or sparks for Fleeting notes. - Extraction: **Select one line** → “Refactor: Extract selection to new note (content only).” - Keep each line short and self-contained. - [F1] Short action — context/https://j.cv/agents-the-new-internet-3-5/ #fleeting #inbox - [F2] Short action — context/https://j.cv/agents-the-new-internet-3-5/ #fleeting #inbox - [F3] Short action — context/https://j.cv/agents-the-new-internet-3-5/ #fleeting #inbox ---