Launch While It's Still Embarrassing

For years, founders have repeated a Reid Hoffman line back to each other: if you're not embarrassed by the first version of your product, you've launched too late. It used to be brave-sounding advice you read on a panel. Now it describes the working day.

You can, if you want, launch twenty MVPs in a single day. The bones look production-ready for at least the first twenty minutes someone spends with them. The friction between I have an idea and there is something on the public internet that does it has collapsed to the cost of an evening and a few dollars of API spend. The advice hasn't changed. The cost of following it has dropped by two orders of magnitude.

My tools, our tools

If you can build twenty things a day, nineteen of them have no business being shared with anyone else. They are tools for you. The list of things one person can now build for their own use, that nobody else will ever touch, has just got enormous. Spreadsheet replacements you don't have to learn Excel for. Calculators specific to the way you actually think. A personal version of a vendor's website that fixes the part you hate. Things that would have been overkill a year ago, because the cost of building them was higher than the value of using them. That arithmetic has flipped.

Those tools are yours. They live on your machine. They don't need an audit log or authentication. Someone else's data isn't in them. They have one user, and they will probably outlive their usefulness by Tuesday.

A small fraction will need to do more than that. Someone else will use it. A customer's data will go through it. Two people will edit it at once. A regulator will eventually ask about it. The moment a tool stops being a single-person artefact, the discipline around it has to change completely. The contract for my tool and the contract for our tool are not the same contract.

That distinction matters more, not less, when shipping is cheap. The cost of putting an our tool into the world with my tool discipline doesn't go down just because you got there in a half-hour.

The discipline isn't new

The disciplines that keep the our tools side from blowing up are not novel. Data model first. State machine underneath. Read what the agent wrote before you ship it. Don't extend the stack past the part you understand. None of it was invented by AI. It was sitting in plain view for thirty years, occasionally written down in style guides that nobody read after onboarding.

AI didn't create the problem of it only works on my machine. It just made it more prevalent. What's actually new is that the discipline can be installed into an agent as a file: a CLAUDE.md, an AGENTS.md, or an editor's equivalent. The transmission cost has collapsed from months alongside a junior to a single curl. People who get the ceiling raise are the ones who already had the discipline and now have it as an installable artefact. People who get a faster floor and no ceiling are the ones who didn't, and won't.

What I built

Two repositories sit on the two sides of this for me. Both public, both MIT. Flatpack is a standard for my tools: a single HTML file you can double-click, structured so a different agent can read it cleanly six months from now, carrying a small inline manifest that knows what would trigger promoting it. Baseplate is a starter template for our tools: a small full-stack foundation with the boring decisions already made, designed for agents to extend rather than invent. The deeper writing on both is in the repos. They are not the point of this post.

Launch while it's still embarrassing

If you can launch twenty things a day, two readings of the old advice are now available. The lesson might be launch more, launch faster, volume is the strategy. It might also be the bar has been quietly raised, and the embarrassing version is now closer to what we used to call good enough. I don't know which one is right. I don't know that any of us will know until we see what the market actually rewards over the next year or two.

What I do know: in either world, the cost of putting your first non-you user in front of a bad our tool goes up, not down. The pronoun is doing real economic work. The free move available regardless is to install minimum-quality discipline on the our tool side. It won't shoot you in the foot. If the recipe that gets you what we cheerfully call scale turns up later, the bones are there. It won't hold you back.

The advice is the same it has been. Launch while it's still embarrassing. Just be honest about which kind of embarrassing it is.