Specs

"PRD is dead" — fine. Then how are you handling scope drift?

The prototype is the spec, they say. Okay — show me the prototype for a rate-limit policy, and tell me which version everyone agreed to.

The Cadenly TeamUpdated June 30, 2026

You've heard it by now. AI moves a team from idea to prototype so fast that the PRD looks like a relic. Why write a ten-page spec nobody reads when you can put a working prototype in front of stakeholders by lunch? The prototype is the spec. Documentation is dead.

Take it seriously, because the speed is real. But a PM posted the sharper version recently: they keep hearing PRD is dead, yet the people declaring it never explain how they handle scope drift, versioning, non-UI features, or platform work without one. That's the right question, and it almost never gets answered.

The prototype answers "what could this look like." Not the rest.

A prototype is great at one job: aligning people on a feel. But walk through what it can't hold:

  • Scope drift. A prototype is a snapshot of one moment's thinking. When the decision changes next week, it either gets rebuilt or silently goes stale — with no record of what changed or why.
  • Versioning. "We decided X, then changed to Y after the partner call." A doc carries that. A Figma file shows you the current state with no memory of how it got there.
  • Non-UI features. Show me the prototype for a rate-limiting policy, a data-retention rule, a billing edge case where the charge differs by 1%. A huge amount of real product work has no screen.
  • Platform and dependency work. Which edge cases, which failure modes, which downstream system this touches — none of that renders. It has to be written, or discovered in production.

The real failure mode isn't "no document." It's "no source of truth."

Here's what the death-of-the-PRD crowd gets right and wrong at the same time.

Right: the 10-page narrative PRD that nobody reads is dead, and good riddance. If it sits in Confluence while the team works off Jira and Slack, it was never the source of truth anyway. It was theater.

Wrong: killing the artifact doesn't kill the need. The need is a single place where the current agreed answer lives, edge cases included, that updates when decisions change and that engineering can trust. If you don't have one, you don't have a faster process — you have a process that lost its memory.

What replaces the dead PRD

  • A living spec, not a frozen document — it updates when the decision updates and keeps the history.
  • Structured stages, not one giant doc — flow, gaps, requirements, test cases, each reviewable on its own.
  • Coverage for the invisible — the non-UI rules and platform dependencies a prototype can't show.
  • A human gate at each stage — because speed with no checkpoint is just scope drift at a higher frame rate.

So is the PRD dead? The old artifact, mostly yes. The function it served, absolutely not. AI made the prototype cheap, which exposed how little value the document was adding — but it didn't touch the need for a trustworthy, versioned, edge-case-aware source of truth. Moving faster makes that need more acute, not less. The people swearing PRD is dead have usually just stopped writing the bad version — and quietly stopped answering the original question.

Key takeaways
  • The 10-page PRD nobody reads is dead; the function it served isn't.
  • Prototypes can't hold scope drift, versioning, non-UI rules, or platform work.
  • The real need is a single trustworthy, versioned source of truth.
  • Replace the doc with a living, staged spec plus human gates — not a prototype.

Try a living, staged spec in Cadenly

Cadenly replaces the dead 10-page PRD with a staged spec — flow, gaps, requirements, test cases — that versions itself as decisions change.

Start free →