Why spec compliance matters right now

AI coding agents have become fast enough that reviewing their output is now the bottleneck. A recent Show HN thread on "Specification First Agentic Development" drew 34 points and 31 comments — a comment-to-vote ratio near 0.91, which is unusually high. That ratio tells you something important: people were not simply upvoting a neat idea. They were arguing about it. And the argument was not about whether specs are good. It was about whether anyone can actually prove that an agent followed the spec it was given.

Today's tooling does not answer that question. Copilot, Cursor, and Claude Code help you write code faster, but none of them audit their own output against your original requirements. PR review bots like CodeRabbit and Greptile check whether code is buggy, well-structured, or idiomatic — they do not check whether it delivers what you promised your team, your customer, or your auditor. The result is a gap that engineering managers feel personally: they sign off on AI-written pull requests, and when something goes wrong, they are the ones answering for it without a single piece of evidence that the spec was ever met.

The timing matters because the buyer already exists. Teams using agentic coding tools in 2026 have engineering managers with real budgets and real accountability. The vocabulary — "spec-first", "agentic development", "compliance" — is already in circulation and has not yet been productized by a major vendor. That window is narrow. SpecProof is built to fill it with the smallest possible surface: two text boxes, one report, and an archivable hash that lets you prove, after the fact, exactly what was checked and when.

How it works

Paste your spec

Drop in a Markdown spec with numbered items like S1, S2, S3. Each line becomes a checkable requirement, so the report can point to a specific clause rather than a vague impression of what was supposed to happen.

Paste your agent's diff

Add the patch or git diff your agent produced. SpecProof reads the changes and cross-references them against every spec item, capturing the exact file and line that satisfies — or fails to satisfy — each requirement.

Get a compliance report

Receive a single-page report with a summary line, a per-item table of satisfied, partial, and no-evidence verdicts, scope-creep warnings, and a timestamped hash you can archive for incident review.

What you get

Line-by-line verdicts

Every spec item is marked satisfied, partial, or no evidence, with the exact file and line that proves it. No hand-waving, no "looks good to me" — just a table you can defend in a review.

Scope-creep detection

We also flag behavior your agent added that your spec never asked for. Silent feature creep is the most expensive kind of drift, and it is invisible to any review tool that only looks at code quality.

Archivable evidence

Each report carries a timestamp and a hash of both the spec and the diff, so you can paste it into an incident review or a compliance folder and prove exactly what was checked, and when.

Prove your AI wrote what you asked for.