AI Agents Favor Vertical Slices Over Platform-First Architecture

For decades, engineering teams justified building horizontal platforms before shipping features, since boilerplate was costly and infrastructure investment paid off over a long roadmap. But research shows roughly 60% of planned features become obsolete before they ship, meaning that heavy upfront investment often locks teams into building the wrong thing.

AI agents make this problem impossible to ignore. McKinsey's 2025-2026 research highlights an 82-point gap between AI adoption (88%) and organizations seeing real bottom-line impact (6%), while a newly described '24-hour sprint' model shows agents can take a spec to tested code overnight when the work is small and clearly bounded. Horizontal, cross-cutting platform work instead gives agents sprawling context and no clear finish line — the opposite of what they need to operate reliably.

The alternative, drawn from Jimmy Bogard's vertical slice architecture, organizes code by feature end-to-end rather than by technical layer. Each slice becomes a self-contained unit an agent can build, test, and ship independently, which explains why organizations like Medtronic — operating under strict FDA oversight — are adopting the pattern to keep agent output verifiable.

The SDLC doesn't disappear; it compresses. Specs become explicit contracts agents implement against, review splits into fast spec-verification and slower value-verification, and human judgment remains the check on ambiguity and accountability. The unresolved question is what stops teams from producing endless slices without ever shipping something real — a governance problem more than a technical one.