Ship-in-Research-Preview Speed Loop
Cut idea-to-user time from months to a week by branding launches as previews and pre-wiring the launch chain
- Difficulty
- Moderate
- Time to result
- ~weeks to results
- Steps
- 4
- Confidence
- 92%
Anthropic's Claude Code team compresses shipping timelines from six months to one week (sometimes one day) by lowering the commitment and friction around each launch. Features go out branded as 'research preview' so the team owes users nothing permanent, and a standing cross-functional process means any engineer can trigger a marketing/docs launch the next day without negotiating. The PM's job is to build this machine, not to gatekeep each release.
Origin
Cat Wu's practice as Head of Product for Claude Code at Anthropic, contrasted against pre-AI PM norms of multi-quarter roadmap alignment.
Core principles
- 01Speed is now the core job of the PM, not roadmap coordination
- 02Reducing commitment reduces the cost of shipping — a preview is reversible
- 03Pre-negotiated cross-functional processes remove per-launch friction
- 04Dogfood internally before the preview goes out
- 05Buggy launches are survivable if they don't block the core use case, because feedback arrives fast
How to run it
- 1
Brand it as a research preview
Ship almost all features clearly labeled as an early, experimental product that may not be supported forever. This signals to users it's an idea being tested and lowers the team's commitment bar so something can go out in a week or two.
Pro tip Explicit 'this is a preview' branding is what psychologically frees the team to ship before it's polished.
- 2
Dogfood internally first
Engineers ship only after the feature has been used internally and they judge it ready. Internal use is the quality bar that replaces heavyweight external QA gates.
- 3
Post to a standing launch room
When ready, the engineer posts the feature in an 'evergreen launch room' where docs, PMM, and DevRel leads are already watching and can turn around the marketing announcement the very next day.
Pro tip The tight, always-on process is what lets any engineer ship without negotiating each time — the PM builds this once.
- 4
Collect feedback and fix in the next release
Accept that some previews will be rough. Because feedback arrives quickly, defects that don't block the core use case get fixed in the next iteration rather than gating the launch.
Watch out Only tolerable when the buggy feature is not blocking your top user's core job — that line is non-negotiable.
In the wild
An observer built a calendar of Anthropic launches and found a major feature or product nearly every day. Cat attributes this less to having the best model and more to being 'very low on process' — every person feels empowered to take an idea to the world in under a week.
→ Sustained daily-launch pace across the org for several quarters.
Common mistakes
Treating every launch as a permanent commitment
Framing features as finished products raises the internal bar so high that shipping slows to a quarterly cadence. Research-preview branding removes that weight.
Renegotiating the cross-functional handoff every time
If marketing, docs, and DevRel involvement is bespoke per launch, friction kills the pace. The process must be standing and pre-agreed so any engineer can trigger it.
Is it for you?
Best for
PMs and eng leads at AI-native product teams who need to ship weekly and are bottlenecked on process and cross-functional alignment
Not ideal for
Heavy-infrastructure projects that genuinely take months, or regulated products where an unpolished public release carries real user harm
From the transcript
“what we do is we actually ship almost all of our features in research preview. We clearly brand this um when we ship something so…”
“when engineers have a feature that they feel is ready and that we've dogfooded internally, they post it in our evergreen launch room.”
“We want to remove every single barrier to shipping things.”
“Launching a feature that is buggy is the kind of thing that would have kept me up at night, but it is something that I'm…”
From the episode
How Anthropic’s product team moves faster than anyone else
Cat Wu (Head of Product, Claude Code)