The requirements document you wrote is becoming the company's costliest waste

Prototype-first beats PRD-first

Product requirements documents are under siege. According to Huxiu (虎嗅), a product team that spent three weeks debating a tagging feature used Anthropic’s Claude Code to spin up a clickable prototype in two hours, then learned from five user tests that nobody cared about the debated detail — they wanted visibility into processing progress. Do teams still need to spend six weeks on research, a 20‑page PRD and months of engineering queue time when a rough prototype plus quick user interviews can surface the real problem in hours?

Big teams are changing how they work

It has been reported that leading AI firms and platform companies are already rewriting playbooks. Anthropic’s product group has reportedly shifted to live prototype-driven research; it has been reported that Claude Code’s annualized revenue exceeds $2.5 billion, and Zapier has reportedly put AI tools in the hands of most employees to speed iteration. Even Rakuten engineers have reportedly used agentic tools to operate autonomously on massive codebases. The shared conclusion: execution is no longer the sole bottleneck — fast, cheap prototyping is.

Implications for PMs, orgs and geopolitics

That does not mean PRDs are dead. For complex system logic, compliance, and cross‑departmental coordination, formal documentation remains essential. The change is about sequencing: validate with cheap, clickable prototypes first; document what’s been proven afterward. But this new loop assumes ready access to advanced generative tools — an assumption that could be affected by export controls, U.S.–China tech tensions and corporate policy on third‑party models. Will Chinese teams be able to replicate Silicon Valley’s rapid loop if access to cutting‑edge models is constrained?

Verdict

The cost of a bloated PRD is no longer just wasted hours; it is delayed learning and missed product‑market signals. AI makes prototypes cheap. What remains scarce is judgment: knowing when a prototype is “good enough” to ship and when to stop iterating. The companies that figure out that balance — and align documentation to validated truths rather than assumptions — will move fastest.