Recruit Like It Costs You Something
Graham's core move is manual, personal recruitment: go find your first users one by one, treat every conversation as irreplaceable signal, and resist the urge to let a growth loop do it for you. For an AI-native founder, this is even more urgent. When anyone can spin up a comparable product over a weekend, the moat is not the build — it is who trusts you and why. That trust is not programmed; it is earned face to face before automation has any face to show.
Delight Before You Delegate
The "surprisingly good" customer experience Graham describes — the one that creates genuine lovers, not just users — is exactly the signal you need before you hand work to agents. You cannot write a good eval for a workflow you have never done yourself. Serving the first fifty customers by hand teaches you what good output actually looks like, which is the only honest foundation for a feedback loop that keeps agents on track.
Manual Work Is System Design in Disguise
Every tedious thing you do for a customer in month one is a workflow waiting to be specified. Graham frames the unscalable phase as market research disguised as customer service; for an AI-native founder, it is also system design disguised as manual labor. The tasks, the edge cases, the moments where judgment was required — those are the inputs to task definitions, tool lists, and the permission boundaries that determine whether an agent handles the next thousand customers well or badly.
Founder-Led Sales Stays Non-Negotiable
Until you have closed deals yourself, you do not know why someone actually bought, and you cannot design an agent-assisted sales motion around a value proposition you learned secondhand. Founder-led sales is not a phase to exit quickly; it is where you discover the real job-to-be-done and confirm that your domain expertise is the unfair advantage that gets your email opened at all. Agents can scale the motion. Only you can discover it.
