# Effective Verification for Humans and Agents > [!question] Lingering Questions > - Is "AI-assisted" too coarse a metric to be useful for cross-team or cross-project comparison, and if so, what's the right taxonomy (autocomplete / agentic-with-review / full spec-handoff)? > - What replaces line-level code reading as the mechanism by which junior engineers build professional judgment and "taste"? > - How do teams in regulated/legacy environments capture the review-time savings the panel describes without "nothing goes live without human eyes" becoming the new bottleneck? > [!tip] Article Ideas > - "Oracle Driven Development for WordPress" — mapping Martin Davidson's build-the-source-of-truth-first pattern onto WordPress's existing PHPUnit/property-testing culture for Core/plugin review teams. #articleseed > - Cross-reference Erin's "multiclass system" concern (legacy/regulated teams can't capture agentic speed gains) against The WPOCC's equity-in-tooling-access discussions. #articleseed > [!link] Cross Reference > - [[3. Resources/Permanent Notes/External Test Oracles Make AI-Generated Code Trustworthy]] — same "verification needs an independent check, not just review" territory, from a different source (Phargo/PHP interpreter case) > - [[3. Resources/Permanent Notes/Delegating a Decision to AI Does Not Transfer Accountability for It]] — already captures this panel's "own the output" accountability claim from a different source > - [[3. Resources/Permanent Notes/Cheap Generation Raises the Value of Deterministic Verification]] — the type-system/deterministic-guardrail argument (Rust) parallels this note's economic framing > - [[3. Resources/MOCs/AI & the Future of Knowledge Work MOC]] > - [[3. Resources/MOCs/AI Agent Safety MOC]] > [!quote] Vocabulary > - **Oracle Driven Development** — Martin Davidson's term for establishing an authoritative source of truth (tests, reference implementations, benchmarks) before writing code, then progressively trusting agents as spec and design solidify. > - **Blast-radius review** — reviewing what an agent's change actually touched/affected via a natural-language summary, rather than reading the diff line by line. > - **MC/DC testing** — modified condition/decision coverage; a safety-critical-software-grade test-coverage standard the panel frames as becoming ordinary once agents can run it constantly. ## Bib capsule LeadDev (panel, sponsored by CodeRabbit), "Effective verification for humans and agents," YouTube, August 13, 2026. https://www.youtube.com/watch?v=_FSEkCBJZls ## Author's thesis (Q1+Q2) Code review's old purpose — catching human mistakes line by line — no longer maps cleanly onto AI-generated code, which fails structurally and architecturally rather than through typos. The panel's converging answer is not "review less" but "review differently": build independent oracles (tests, fuzzers, mutation/property testing, deterministic type-system guardrails), read agent-authored natural-language summaries instead of diffs for structural catches, and keep a single named accountable human for anything that ships — even when that human never reads a line of code. ## Key arguments (≤3) 1. **Verification infrastructure replaces line-by-line review as the real check.** Oracle Driven Development — establish tests/reference standards/benchmarks before an agent writes code, then progressively hand off trust — plus mutation/property/fuzz testing absorbs the correctness-checking work manual review used to do. 2. **Structural/architectural review can happen above the code level.** The panel's strongest empirical example: an agent silently added an unrequested middle tier during a 40-hour unattended run, caught not by reading a diff but via a natural-language "what changed and why" summary. 3. **Accountability doesn't automate away — and neither does its unevenness.** "AI made it. You own it" fixes who is responsible; but legacy/regulated teams (banking, insurance) can't adopt the greenfield oracle-first approach, creating what one panelist calls a "multiclass system" where only nimble teams capture the benefit. ## Claim atoms ### Oracle Driven Development replaces line review with source-of-truth-first design > [!quote] ID:rw_1 > I like the term Oracle driven development... you start off with your oracle. How are you going to test this thing? Where/why: ~25:57, Martin Davidson explaining his verification methodology — build the test/reference standard first, hand off trust progressively, rather than reviewing output after the fact. ### Reviewers should read agent-authored change summaries, not just diffs > [!quote] ID:rw_2 > You can look at what happened at a high level using natural language and it will explain to you exactly what went on... I don't need to look at the code to catch it. Where/why: ~39:48, David's catch of an agent-added, unrequested architecture tier — concrete counter-evidence that dropping line-by-line review doesn't mean losing the ability to catch structural problems. ### Legacy and regulated codebases can't capture agentic coding's speed gains equally > [!quote] ID:rw_5 > They're going to spend more of their time undoing what Claude has written because it doesn't work with our stack... it's like a multiclass system in the tech world. Where/why: Erin's rebuttal to Martin's greenfield-optimized account — flags an equity/access gap in who actually benefits from agentic coding. ### Strongly-typed languages act as deterministic guardrails alongside AI review > [!quote] ID:rw_4 > Rust basically gives you a lot of guard rails for free... a whole class of bugs that you know you can't have. Where/why: ~29:21, Martin explaining why his team writes almost exclusively in Rust for agent-authored code — a deterministic complement to probabilistic AI review. ## Permanent candidates - The oracle-first verification claim (H1/H2) and the "own the output" accountability claim (H3) are already captured by [[3. Resources/Permanent Notes/External Test Oracles Make AI-Generated Code Trustworthy]] and [[3. Resources/Permanent Notes/Delegating a Decision to AI Does Not Transfer Accountability for It]] respectively. No duplicates created. - **New**: the "multiclass system" claim (H5) is not yet captured — promoted below as [[3. Resources/Permanent Notes/Agentic Coding Speed Gains Are Unevenly Available Across Codebases]]. - **New**: the natural-language-summary-over-diff review claim (H2) is distinct enough from the oracle-testing claim to promote separately — promoted below as [[3. Resources/Permanent Notes/A Change Summary Can Catch Structural Problems a Diff Review Would Miss]]. ## Processing checklist - [x] Checked for existing Literature Note — none found - [x] Checked for existing Permanent Notes covering this source's claims — found two already-covered claims (oracle-first, accountability), did not duplicate; promoted two new ones (multiclass system, change-summary review) - [x] Connected into [[3. Resources/MOCs/AI & the Future of Knowledge Work MOC]] and [[3. Resources/MOCs/AI Agent Safety MOC]] ## Links - [[3. Resources/MOCs/AI & the Future of Knowledge Work MOC]] — verification/review-role redesign under AI-authored code volume - [[3. Resources/MOCs/AI Agent Safety MOC]] — accountability and deterministic-guardrail themes