Public transcript

Everyone’s shipping more. Does any of it matter? | Claire Vo

By: Lenny's Podcast
Length: 24 min
Language: English
Text from: the video's captions
Transcribed: 2 October 2026

Read the full transcript Open on YouTube

Notes in English.

Summary

Claire Vo, speaking at Lenny's Summit, argues that AI has made execution nearly limitless while conviction about what to build has become the scarce resource, making traditional feature road maps dangerous. She shares her own experience of shipping more than ever while doubting any of it mattered, and warns of three traps: the backlog trap, the parity trap, and the churn trap. She proposes replacing road maps with durable convictions, upfront evidence definitions, and a classification of ships as probes, experiments, or promises. Her closing ask is for the audience to write their last road map and play the ambition game next year rather than the velocity game.

Key points

Claire Vo says execution capacity has outrun her ability to discover meaningful, commercializable products, and she is out of good ideas despite shipping more than ever.
Engineering capacity used to be the scarce resource, which made road maps and prioritization work; now the bottleneck has moved from building to conviction about what is worth building.
She warns of three traps: the backlog trap (AI builds everything on the list), the parity trap (competitors all arrive at the same obvious product), and the churn trap (shipping, abandoning, and never compounding learning).
Road map zero means every visible feature is plausible and buildable, so buildability and effort stop being meaningful proxies for what matters.
AI cannot turn an untested assumption into fact; you still need real customers, real data, repeated tests, and a unique point of view.
She advocates durable convictions with disposable features, distinguishing good stubborn (staying with the problem, revising the solution) from bad stubborn (moving the goalpost because you have tokens).
Not every ship is a promise; be honest about whether something is a probe, an experiment, or a promise, because code is abundant but customer trust is not.
Next year is the ambition game, not the velocity game: measure how many huge experiments you run per month rather than PRs or revenue per headcount.

Questions it answers

00:00Opening confession

What is Claire Vo's big confession about shipping with AI?

Despite having the best coding agents, tools, and 40 Grock bots, she is out of good ideas: execution has outrun her ability to discover meaningful, commercializable products.

She can build almost anything but lacks conviction about what is worth building.
The bottleneck has moved from building capacity to conviction.

06:06The product graph story

What happened when she built the product graph for ChatPRD?

She built a nearly one-shot product graph for ChatPRD that matched competitors feature-for-feature, but felt it belonged in the trash because it was not differentiated and the ROI was shaky.

Shipping tokens did not test or strengthen her conviction.
Her AI factory automated fixes and tech debt but she is not sure any of it mattered.

10:08Three traps

What three traps does limitless AI execution create?

The backlog trap (AI builds everything on the list), the parity trap (everyone builds the same obvious product), and the churn trap (abandoning ships without learning or compounding).

All three feel productive from the inside.
Together they accelerate your path to mid.

12:32Road map zero

What is road map zero and why does it kill the road map?

When every visible feature is plausible and buildable, effort stops being a proxy for what matters and prioritization stops being strategy, making feature-and-date road maps dangerous.

An AI factory plus an old road map reaches the three traps at machine speed.
The remaining constraint is truth, not code.

14:05Convictions over features

What should replace the feature road map?

Build durable convictions, define evidence upfront that would prove them true or make you stop, keep the factory to make fast contact with reality, and hold durable convictions with disposable features.

Good stubborn stays with the problem and revises the solution; bad stubborn moves the goalpost because you have tokens.
AI cannot turn an untested assumption into fact.

17:41Probes, experiments, promises

How should teams think about commitments to customers?

Not every ship is a promise: classify work as probes, durable experiments against convictions, or promises customers can rely on, because code is abundant but customer trust is not.

She kept the product graph feature-flagged off to avoid burning customer trust.
Be honest about the strength of conviction behind each feature.

20:22The ambition game and the ask

What game should product teams play next year?

Next year is the ambition game: take huge swings and measure how many big experiments you run per month, then write your last road map — one of ambition and evidence rather than feature lists with dates.

OKRs should count huge experiments per month, not PRs or revenue per headcount.
Be big on ambition and fuzzy on the specifics.

Notes

The confession

Claire Vo, a product leader of over two decades, says she is shipping more than ever with the best coding agents, tools, and customer context, yet she is "out of good ideas."
Execution has outrun her ability to discover meaningful, "commercializable" products; her bottleneck moved from "what can I build" to "what do I believe is worth building."

Why road maps made sense before

Engineering capacity used to be the scarce resource; PMs prioritized and said no, engineering pushed back on scope, design asked to wait.
Scarcity filtered out bad ideas because items "below the cut line" never shipped.

The product graph story

She built a "product graph" for ChatPRD: an insights engine, semantic product graph, auto wiki, agent-consumable, nearly a one-shot, feature-for-feature matching competitors.
She felt it "belongs in the trash" because it was competitive but not differentiated; the ROI was shaky and the interface questionable.
Meanwhile her AI factory ran itself: automated customer fixes, obliterating tech debt (rearchitected ChatPRD 70 times), no issue tracking, just shipping PRs — and she is not convinced any of it mattered.

Three traps

Backlog trap: AI builds every item, but clearing requests is not meaningful progress.
Parity trap: competitors talk to the same customers and build the same obvious product.
Churn trap: ship, see noise, abandon, never learn or compound.

Road map zero

Every visible feature becomes plausible and buildable, so buildability and effort stop being proxies for what matters; prioritization (e.g., RICE) stops being strategy.
An AI factory plus an old road map gets you to the three traps at machine speed.

What to do instead

Build durable convictions; define upfront what evidence would prove them true or make you stop.
Keep the factory, but place it correctly in the process; make faster contact with reality.
Be good stubborn (stay with the problem, revise the solution), not bad stubborn (move the goalpost because you have tokens).
Classify ships as probes, experiments, or promises; code is abundant, customer trust is not — she kept the product graph feature-flagged off to avoid burning trust.

The ambition game

The last 12–18 months were the velocity game; next year is the ambition game: huge swings, big experiments in two or three weeks.
For OKRs, ask how many huge experiments you run per month, not PRs or revenue per headcount.
Ask: build your last road map — one of ambition, evidence, and conviction, not feature lists with guessed impact and dates.

Text from the video's own captions. Summary and notes written by AI from the transcript, so check anything important against the recording.