The Noticing Stays Human
Graham's central move is to shift the locus of idea generation from imagination to observation. You live with a problem, you feel its friction, and only then do you build. That instinct becomes more important, not less, when prototyping collapses from months to hours. When any competent founder can spin up a working agent-driven workflow in a day, the scarce resource is no longer the ability to build — it is the judgment to know which problem deserves to be built toward. The bottleneck moves from typing to thinking, and thinking starts with noticing.
Small and Desperate Before Broad and Useful
Graham is insistent that the right early market is a small group that wants something badly, not a large group that might find it convenient. For AI-native products, this maps directly onto the design question of where humans remain in the loop. A narrow, desperate user has specific needs, sharp feedback, and low tolerance for failure — exactly the conditions that force founders to build real evals and honest feedback loops rather than shipping vague autonomy and hoping it holds. The desperate early user is your first and most honest harness.
The Edge of the Newly Possible
Graham points to ideas that only became viable recently as the richest territory. Software that can be directed in natural language is precisely such a threshold. The founder who recognizes that an entire workflow — one that previously required a trained specialist and three software tools — can now be orchestrated by an agent with good context and clear permissions is standing at that edge. The idea looks like a toy to outsiders because the category is new. That is the signal, not the warning.
Taste Decides What Gets Built
Once you have noticed the problem and grasped what is now buildable, the remaining constraint is taste — the ability to recognize when the output is good and when the agent is confabulating competence. Graham's editor instinct, the person who can read a draft and know whether it solves the problem, is the same instinct a founder needs when reviewing what an agent produces. Distribution and trust do not emerge from the model; they are designed in by a person who knows what the work is supposed to accomplish and can tell the difference.
