# Agentic Coding Speed Gains Are Unevenly Available Across Codebases ## Claim Greenfield, unregulated codebases capture agentic coding's speed gains far faster than legacy, regulated ones, because the fastest verification patterns (oracle-first design, near-total trust handoff) require a freedom to redesign the codebase and its review process that regulated environments don't have. <!-- atomicity: 1 idea ✓ --> ## Scope & conditions - Holds when the constraint is structural — an existing legacy stack, compliance requirements, or an install base that can't tolerate the churn of a redesigned verification approach. - Does not claim regulated teams get zero benefit from AI-assisted coding — only that they can't capture it at the same rate as teams free to redesign around oracle-first, low-review workflows. - Breaks down if a regulated team can carve out a genuinely greenfield component (new microservice, new module with no legacy dependents) inside an otherwise legacy system — the constraint is about the codebase's freedom to change, not the industry label. ## Evidence On a LeadDev/CodeRabbit panel, Martin Davidson (CTO of a two-person consultancy) describes writing and reviewing zero code personally, running a fleet of agents against an oracle-first verification process. Erin, an agile coach working inside legacy/regulated environments (banking, insurance), pushes back: teams there spend more time "undoing what Claude has written because it doesn't work with our stack," calling the resulting gap "a multiclass system in the tech world" — the same tooling produces uneven benefit depending on how much freedom a team has to change its verification process, not just its willingness to adopt AI. ## Links - [[AI Agent Safety MOC]] — why the same agent produces very different results depending on where it's pointed - [[3. Resources/Literature Notes/Effective Verification for Humans and Agents]] — because evidence-for: source Literature Note for this claim - [[3. Resources/Permanent Notes/External Test Oracles Make AI-Generated Code Trustworthy]] — because contrasts: names the verification pattern that produces the speed gain this note says isn't equally available - [[3. Resources/MOCs/AI & the Future of Knowledge Work MOC]] — because applies-in: an access/equity gap in who benefits from AI-assisted engineering work ## Tests / falsifiers - Compare AI-assisted code review-time reduction (as % of engineering time) between matched greenfield and legacy/regulated teams at similar AI-adoption levels — if the gap doesn't hold, this claim is wrong. - Find a regulated team that successfully adopted oracle-first, low-review verification without carving out a greenfield component — a genuine counterexample would narrow this claim's scope considerably. ## Maintenance - Last tended: 2026-08-16 - Next review: 2026-11-16 ## Source - [[3. Resources/Literature Notes/Effective Verification for Humans and Agents]] — reports Erin's "multiclass system" rebuttal to Martin Davidson's greenfield oracle-first account.