The Zero-to-One Enclave
Conditions that let a startup-shaped team actually ship inside a big company
- Difficulty
- Advanced
- Time to result
- ~months to results
- Steps
- 6
- Confidence
- 88%
Most big-company innovation labs fail. Martin describes the specific conditions that made NotebookLM the exception: a senior leader with a blunt mandate, a team small enough to fit in one room, deliberately stripped process, permission to use non-native tools, and a public channel to users the parent company would never normally allow.
Origin
Raiza Martin's account of building NotebookLM inside Google Labs (founded ~2021 by VP Josh Woodward), from a 20% project with a single full-time engineer to launch with three engineers, a designer, a PM and Steven Johnson.
Core principles
- 01The mandate must be stated in one sentence by a senior leader: ship AI products and build businesses out of them.
- 02Ambiguity about the mission is fatal; ambiguity about the roadmap is fine.
- 03Fewer processes, maybe even to a fault — the fault is cheaper than the drag.
- 04Roles collapse: PMs, engineers and designers crank on mocks, PRDs and implementation in the same meeting.
- 05Build in public on the platforms your users already live on, not the ones your employer prefers.
- 06Keep the team small for far longer than feels comfortable — NotebookLM launched with three engineers and had fewer than ten a year in.
How to run it
- 1
Get a one-sentence mandate from a senior sponsor
Before joining or founding the enclave, force the sponsoring executive to state the mission in one line. Martin asked Josh Woodward 'what's the mission, what are we here for?' and got: we're going to ship AI products and build businesses out of them.
Pro tip A mandate that names an outcome (build businesses) rather than a domain gives you cover to kill and pivot without renegotiating.
Watch out If the sponsor cannot state it in one sentence, the enclave has no air cover and will be process-absorbed within two quarters.
- 2
Stay radically small
Resist headcount until the product has demonstrated pull. NotebookLM began as one engineer plus part-time contributors, launched as Project Tailwind with three engineers, one designer, a PM and Johnson, and had under ten engineers a year after launch. Headcount arrived only in anticipation of a roadmap earned by traction.
Pro tip Use the anticipation rule: staff for the roadmap you have proven, not the one you hope for.
- 3
Delete process on purpose
Run combined meetings where product, engineering and design write the mocks and the PRD together while implementation is already underway. Accept that you will occasionally get it wrong ('maybe even to a fault') as the cost of speed.
Watch out This only survives if leadership has explicitly pre-approved the deviation from company norms; done covertly it reads as insubordination.
- 4
Claim the right to use the outsider's tools
Ask what you would use if you were building outside the company, then use that. Martin insisted on a Discord server despite colleagues asking why not a Google Meet or a Google Group — the answer being that she did not know how to use those for a community, and Discord was where builders already were.
Pro tip Frame the ask around user habit, not personal preference: this is where the community already is.
Watch out Expect internal friction on tooling before you get any external friction on the product.
- 5
Build in public and stay in the channel
Ship constantly, tweet daily, and be personally present in the community channel every day — reading everything even when not replying. The Discord grew from a launch fear of 'what if nobody joins' to roughly 60,000 members.
Pro tip Presence beats response rate. Martin reads everything and says so publicly, which is itself a retention mechanism.
- 6
Sequence business after interestingness
Break the mandate into steps: build something interesting first, then go figure out the business. Once traction is real, lean on the parent company's distribution, monetisation and commercial machinery instead of reinventing it.
Pro tip The enterprise pull signal is unmistakable: employees using your consumer product on personal accounts, and their employer calling you to make it official.
Watch out Deepening the user experience and commercialisation run in parallel — do not let business development consume the shipping capacity that created the pull.
In the wild
NotebookLM started as a 20% project ('talk to small corpus') with one engineer full-time and everyone else chiming in. It launched as Project Tailwind with three engineers, and only added significant headcount roughly a year later once the roadmap justified it.
→ Rising daily, weekly and monthly retention, a 60,000-member Discord, and enterprise inbound so heavy Martin had to hire a business development person to protect her shipping time.
Martin pushed for a Discord server against internal confusion ('why not a Google Meet? why not a Google Group?'), and her main fear at launch was that nobody would join.
→ The server reached about 60,000 members and became the team's primary daily feedback channel.
Common mistakes
Standing up a futuristic side-team without a stated mandate
Lenny notes that big companies repeatedly create off-to-the-side innovation teams and they rarely work. The difference here was a senior leader explicitly naming what the team would do and which normal processes it would skip.
Scaling the team before the product has pull
NotebookLM stayed under ten engineers through launch and its first viral year. Adding people earlier would have created the coordination overhead the enclave exists to avoid.
Insisting on native corporate tooling for community
Defaulting to the company's own products for user communities puts you where your users are not. The Discord worked precisely because it was where builders already lived, not where the parent company preferred to host them.
Is it for you?
Best for
Product leaders founding or joining a zero-to-one team inside a large company, with a senior sponsor and permission to break process
Not ideal for
Teams accountable to quarterly business outcomes on an existing revenue line, or any group without genuine executive air cover for skipping process
From the transcript
“in Labs we have the environment to do it we have the environment to move fast we have far far far fewer processes maybe even…”
“when I joined my mandate was to build a business and I was like all right if we break down the steps here along the…”
“I was like Hey I want to have a Discord right like if we were building outside of Google for sure we would have one…”
“actually notebook started as a 20% project as things kind of do at Google as they they used to”
From the episode
Behind the product: NotebookLM
Raiza Martin (Senior Product Manager, AI @ Google Labs)