VerifiedSpecDrivenDevelopment
title: Verified Spec-Driven Development radar_quadrant: Techniques radar_ring: Assess radar_position: inner
Verified Spec-Driven Development is a technique for directing AI coding agents using specifications that include explicit, machine-checkable acceptance criteria. The agent is considered done only when the verification steps pass — not when it declares the task complete. This tightens the feedback loop between agent output and correct behaviour by replacing open-ended instructions with testable exit conditions.
Standard AI coding agent instructions are underspecified: "implement a login page" leaves the agent to define what done means. The agent finishes, declares success, and a human must manually inspect the output. This review burden scales linearly with agent output volume.
Verified Spec-Driven Development addresses this by extending spec-first thinking into the agent loop. Each spec includes automated checks — unit tests, integration assertions, structured output schemas — that the agent runs against its own work before handing back control. The technique builds on TDD principles but frames them around the agent's self-evaluation cycle rather than the human developer's workflow. Agents with verifiable exit criteria produce more consistent outputs and require less supervisory review.
Placed in Assess at inner position because it is a direct, practical response to a friction point any team running coding agents encounters. The inner position reflects that developers using Claude Code or other coding agents for non-trivial tasks should be evaluating this pattern now. The main investment is upfront spec quality — acceptance criteria must be concrete enough to automate — which raises the bar for prompt writing but pays off in reduced review cycles.
