Your first 30 days with an AI cofounder

Week one

It reads, drafts, and publishes nothing

You give it your site and one paragraph on what you sell. It reads the page, mines the places your customers complain, and starts producing ad concepts. Every one lands in a queue. Nothing is published and no money moves — the point of week one is that you find out what it makes before it can act on anything.

What to do: Read the first batch properly. Kill the ones that sound like a brochure. That is the training signal.

Week two

You start approving, it starts spending

Once you have a feel for the output, set a daily ceiling and let approved creative go live. The ceiling is an authority limit, not a target — it is the amount the system has no path to exceed, which is what makes leaving it running overnight a reasonable thing to do.

What to do: Set the ceiling low enough that a bad week is annoying rather than expensive.

Week three

The first results come back, and most of them are bad

This is the week people quit. Most creative loses — that is true of every ad account, run by anyone. What matters is whether the losses are being cut quickly and whether the survivors are being cloned into new angles. A system that kills the bottom of the pool by morning is working correctly even when the week's numbers look poor.

What to do: Judge the process, not the week. Ask what it killed and why, not whether the number went up.

Week four

You stop reading every card

By now there is enough performance history for the pattern to show: certain formats and hooks keep winning in your niche. That is when most people move from approving each piece to approving a direction and letting it run inside the ceiling — checking the morning brief rather than the queue.

What to do: Move the autonomy dial one notch, not three. It is easy to move back.

How long before an AI cofounder produces results?

Creative appears on day one; meaningful performance data takes two to three weeks, because ads need enough impressions before a kill-or-keep decision means anything. Anyone promising a verdict in a few days is reading noise.

How much time does it take from me?

Most of the commitment is in week one, reviewing the first batches so it learns your taste. After that it is a morning brief — roughly half an hour a week once you stop reading every card.

What should the budget be to start?

Low enough that a bad first month is not painful. The daily ceiling is a hard limit the system cannot exceed, so the sensible move is to start small, watch what week three looks like, and raise it once you have seen it cut a loser on its own.

What happens if it makes something off-brand?

Brand rules — words it will never use, claims it cannot make — are checked on each asset before it reaches your queue. Anything that fails is not shown. In approve-first mode nothing publishes without you regardless.

Can I go back to approving everything?

Yes. Autonomy is a dial, not a one-way door, and moving it back takes one setting. Most people move it up in week four and occasionally step it back down when they change positioning.