Your SDLC Is Your Context Engineering (LeadDev)

# Your SDLC Is Your Context Engineering (LeadDev) > [!info] Machine-mediated deep-read — not the author's words except inside quote blocks. > Ported from an existing Readwise deep-read for doc `01kzp1fhtrv7h8mnrdsnd6xbsy`. Self-reported findings from one month at one small startup — no external audit, no control condition. > [!question] Lingering Questions > - Does the "planning, not code" defect-root-cause finding replicate outside this single dataset, or is it partly an artifact of a review rubric that explicitly steers reviewers to check design-decision alignment first — which would mechanically push blame upstream regardless of where the defect actually originated? > - How does the four-lane risk classification hold up as team size scales past "lean startup" — does CODEOWNERS enforcement on the highest-risk lane become a bottleneck at dozens of concurrent engineers and agents? > - Is the task-level builder/reviewer subagent separation (four isolated contexts per feature) cost-effective at scale, or does it work specifically because the added inference cost is negligible at Vendict's size? > [!tip] Article Ideas > - A WordPress-flavored version of this piece: could Core/CLI/Playground contributing docs adopt a docs/ vs. plans/ vs. adr/ split to reduce AI-agent hallucination on stale/abandoned plans? > [!link] Cross Reference > - [[3. Resources/Permanent Notes/Agent Specification Not Code Is the New Bottleneck]] — same underlying shift (spec/plan quality outranks implementation as the bottleneck), different source and evidence base > - [[3. Resources/MOCs/AI & the Future of Knowledge Work MOC]] > [!quote] Vocabulary > - *Lane* (A–D) — this SDLC's PR risk classification: standard work, spec+plan required, two-approval control-plane, isolated experiment > - *docs/ vs. plans/ vs. adr/* — a three-way documentation split (current merged truth / future active work / permanent superseded-not-deleted decisions) designed so an agent always knows which document to trust ## Metadata - Source: [[3. Resources/Readwise/Articles/Your SDLC is your context engineering]] ## Author's thesis A founding engineer at a small TPRM startup argues that file-level context engineering (AGENTS.md) gives an AI coding agent facts but not judgment — what "done" means, what's off-limits, when to escalate — and that judgment lives in the software development lifecycle itself, so the team rebuilt their entire SDLC (doc lifecycle, risk-based PR lanes, task-level builder/reviewer separation, path-referenced standards, scoped-credential devcontainer) around that thesis. ## Key arguments (≤3) 1. **Test-exemption from PR size caps is a deliberately anti-incentive design choice**: counting tests toward the size limit would train agents to under-test or split tests from implementation — both worse than the cap's original goal. 2. **Builder/reviewer separation must happen at task granularity, not PR granularity**: a single agent session spanning implementation, testing, and self-verification lets the same context grade its own work; four isolated subagents (test designer, failure analyzer, verifier, linter) prevent this. 3. **The article's headline empirical claim — critical issues trace to underspecified planning, not code — is presented as an observed one-month finding but is self-reported, uncontrolled, and possibly an artifact of a rubric that already steers reviewers to check plan/design alignment first.** ## Claim atoms ### SDLC is the agent's real context, not AGENTS.md > [!quote] ID:rw_sdlc-thesis > At Vendict, a lean third-party risk management (TPRM) startup, we rebuilt our engineering process around one thesis: in the agentic era, the SDLC is the context. **Context & use**: Opening thesis — every subsequent structural choice (lanes, doc lifecycles, subagent separation) is presented as a consequence of this claim. **Tags**: #xref  ### Exempt tests from PR size caps to keep agent test coverage high > [!quote] ID:rw_test-exemption > If tests count toward size pressure, the agent learns to either write fewer tests or split implementation from tests into separate PRs. Both outcomes are strictly worse than what the rule was trying to buy. **Context & use**: Design choice #1 — a concrete counter-incentive to guard against when setting AI-agent PR size limits. **Tags**: #xref #articleseed  ### Builder-reviewer separation must live at task level, not PR level > [!quote] ID:rw_builder-reviewer > Builder-reviewer separation has to live at the task level because a single agent session today can be the size of what used to be a full PR a couple of years ago. **Context & use**: Design choice #2 — the mechanism behind why "don't let the agent review its own code" has to apply within a single session's sub-steps, not just across separate PRs. **Tags**: #xref  ### Critical issues trace back to planning, not code > [!quote] ID:rw_planning-root-cause > The critical issues we see come from insufficiently thoughtful planning and underspecified requirements. The real problem is not only how agents write code, but how clearly humans define the work before the code is written. **Context & use**: The article's single strongest empirical takeaway, closing the "lessons of the first month" section — but self-reported, n=1 org, one month. **Tags**: #xref  ## Permanent candidates - **Already captured**: the spec/plan-as-bottleneck claim overlaps [[3. Resources/Permanent Notes/Agent Specification Not Code Is the New Bottleneck]] — not re-created; linked instead. - **New**: builder/reviewer separation must happen at task granularity, not PR granularity — a distinct mechanism claim about *why* self-review fails even across separate PRs when a single session spans the whole task. See new Permanent note below. ## Processing checklist - [x] Connect into the relevant MOC — [[3. Resources/MOCs/AI & the Future of Knowledge Work MOC]] - [x] Checked for duplicate Permanent notes — spec-bottleneck claim already exists, linked not duplicated - [ ] Skim the article's linked DORA 2025 AI-assisted software development report for corroborating data on the "planning, not code" claim ## Links - [[3. Resources/MOCs/AI & the Future of Knowledge Work MOC]] — SDLC redesign for agentic development is core to this MOC's territory - [[3. Resources/Permanent Notes/Agent Specification Not Code Is the New Bottleneck]] — because evidence-for: a second, independent source for the same spec-is-the-bottleneck shift