Based on Lenny's Podcast data

Lenny's Knowledge Sketch · #51

How Anthropic Ships
Faster Than Anyone Else

Cat Wu

Head of Product, Claude Code & Co-work
Anthropic

APR 23 2026

Shipping

Timelines Have
Collapsed

Feature Timeline: Before → Now

Before AI

6 months

Today

1 week

Sometimes

1 day

"We want every person on the team to feel empowered to take their idea from just an idea to out in the world in less than a week."

Ship in research preview, reduces commitment
Evergreen launch room: eng posts → marketing turns around next day
Low process = less friction to ship

PM Role

What Great PMs
Actually Do Now

Less of this

Multi-quarter roadmap alignment
Coordinating partner teams
Writing long PRDs
Waiting for dependencies

More of this

Clear goals + user definition
Concept corner: ship in a week
Cross-functional framework
Building evals

"As code becomes much cheaper to write, the thing that becomes more valuable is deciding what to write."

Skill

Product Taste
is the
Rare Thing

"It is very hard to be the right amount of AGI-pilled."

Build for the current model, not the AGI one
Elicit max capability from what exists today
Guide users to the model's strengths
Patch its weaknesses in the harness

The Eval Habit 10 great evals > 0 evals
Quantify the goal. Track progress. Find the gap.

Model × Product

New Model = Remove Features,
Unlock New Ones

What gets removed

→Prompting interventions the model no longer needs
→To-do list forcing (Opus 4+ does it naturally)
→System prompt sections that are now obvious

"The model will eat your harness for breakfast."

What gets unlocked

→Features built for a future model finally work
→Code review (reliable only from Opus 4.5/4.6)
→Multi-agent parallelism at real scale

Build prototypes slightly ahead of what works, swap in the new model to close the gap.

Vision 1 task

works reliably

Now 6 tasks

multi-coding

Next 50–100

remote Claudes

Stack

When to Use
Which Tool

Claude Code CLI, latest features first, one-off coding tasks
Desktop, frontend + live preview pane; control plane for all sessions
Mobile/Web, kick off tasks on the go, no laptop needed
Co-work, output is not code: decks, docs, Slack drafts, inbox zero

Co-work secret Connect Slack + Gmail + GDrive + Calendar first. Context = quality of output.

Cat's overnight deck: gave it narrative + links → woke up to a 20-page polished draft.

Advice

How to Actually Thrive in
the AI-Driven World

01 Automate the tedious

Every repeated manual task is a candidate. Pass it to Claude. Keep the creative parts for yourself.

02 Get to 100%, not 95%

A 95% automation is not an automation. Put in the elbow grease. Teach it your preferences. Get it reliable.

03 Build apps you use daily

Prototypes that sit unused give you nothing. Real leverage comes from real daily use, iterate until it earns a place in your workflow.

04 Just do things

"Jobs are fake." If you understand the constraints, figure out what you can do and do it quickly. Apologize if wrong. Learn. Move.

"It comes back to product taste. As code becomes much cheaper to write, the thing that becomes more valuable is deciding what to write."

Timelines HaveCollapsed

What Great PMsActually Do Now

Product Tasteis theRare Thing

New Model = Remove Features,Unlock New Ones