Do Things That Don't Scale, Forever
Paul Graham’s most famous advice is do things that don’t scale. Recruit your first users by hand. Do the work manually. Embarrass yourself in strangers’ inboxes. The usual reading treats this as training wheels, an awkward phase you endure until the real scalable machine is ready. I think that reading is wrong, and the agent age is about to make it dangerously wrong.
The standard interpretation goes: do unscalable things early to bootstrap, then automate everything as fast as possible, because automation is the goal and manual effort is waste. Spin the flywheel, stop pushing it by hand. There’s truth in it. But it smuggles in an assumption I no longer buy, that scalable automated work is inherently better than unscalable manual work, and that the manual stuff is something to escape.
Here’s the inversion the agents force. When the scalable work, the automatable, repeatable, specifiable work, gets done by agents for nearly free, it stops being valuable because everyone can do it. Anything scalable scales for your competitors too. If a task can be automated it will be, by everyone, and it collapses to a commodity with no margin. The scalable parts of your business are exactly the parts that won’t be yours for long. Table stakes the moment they’re cheap.
Which means the unscalable things aren’t the embarrassing phase anymore. They’re the permanent edge. The handwritten note an agent could fake but has no real self behind. The genuine relationship built one unhurried conversation at a time. The taste applied personally to each decision instead of batch-processed. The thing made with visible care by an actual human who chose to spend their finite hours on you specifically. None of it scales, and in a world drowning in infinite cheap scalable output, that’s exactly why it’s worth everything. Scarcity moved. It used to live in the scalable. Now it lives in the human.
So I’d rewrite the advice for this decade: do things that don’t scale, and never fully stop. Not a phase. A permanent strategy. Let the agents handle everything that scales. That’s their job, they’ll do it better and cheaper than you ever could, let them. Then spend your own irreplaceable human hours on the unscalable things that are now the only durable edge. Relationships, taste, judgment, care, the personal touch that can’t be mass-produced because the whole value is that it wasn’t.
This connects to earnestness, which Paul Graham also wrote about and which I think ages into the central virtue here. In a world of infinite generated everything, the genuinely earnest human gesture, the thing you actually meant, made by hand, that obviously cost you something real, becomes the rarest and most valuable signal there is. You can’t fake it at scale, which is exactly the point. The fakeable is now free. Only the real is scarce.
The future doesn’t belong to whoever automates the most. The agents win that race against everyone, including you. It belongs to the people who know which human things to refuse to automate. Do things that don’t scale. Then guard them, because they’re all you’ll have left that’s yours.
One of a series of essays. I’m Prajjwal Chittori. prajjwalchittori.com.