The PRD used to be a permission document. Now, it’s a decision record.

Let me explain…

The old flow was slow: Idea → PRD → Design → Build → QA → Ship

The PRD came so early because building was expensive. “Engineering time is the most valuable in the company” was a regular refrain. So you needed permission and buy-in before ever getting anything big built.

The new flow is fast: Idea → 5 prototypes → Evaluate → Kill 4 → PRD for the survivor → Ship

The PRD has moved to the end stages because you don’t need permission to build 5 prototypes anymore. They’re simply an MCP connection to the code base, a configured design system, and a prompt away.

By virtue of moving later in the process, what goes in has changed, too. The best PRDs now are basically what a prototype can’t communicate:

1. The opportunity framing

2. The scope, boundaries

3. Success measurement

4. The behavior contract

5. The rollout plan

6. Risk mngmt

Essentially, the prototype shows what. And the PRD shows why, how you’ll measure, and when you’ll roll back.

In some teams, the PRD has even become optional. Anthropic's Claude Code team writes zero PRDs because the team aligns through conversation and working software.

Most teams won’t go that lean, and that’s because they’re not building a product for themselves, operate in a regulated industry, or have many stakeholders to align. That’s completely fine. Not everyone has to build like Anthropic.

But everyone has to embrace new ways of working.

Here's how to go deeper on them:

1. AI PRDs (full guide + template): news.aakashg.com/p/ai-p…

2. AI prototyping tutorial: news.aakashg.com/p/ai-p…

3. Taste at Speed guide: youtube.com/watch?v=Gp6…

4. Prototyping tool faceoff: youtu.be/74kamD2jUng

5. Claude code scaffolding: news.aakashg.com/p/pm-os

PRDs aren't dead. They've just grown up.