cofounderHow much autonomy to give an AI cofounder
Autonomy is a dial with three positions, and the honest answer is that you should start at the strictest one. Not because the system is untrustworthy, but because you cannot yet tell whether it is — and the fastest way to find out is to read a week of work it wanted to publish and see how much you would have changed.
The three modes
What each one permits
Draft only
Week one, and any time you change positioning
Researches, writes and renders. Publishes nothing, spends nothing.
The point is to find out what it makes before it can act on anything. Most people discover in the first two batches that their brand rules were vaguer than they thought, and fixing that here costs nothing.
Approve first
The default, and where most people stay for a month
Queues a daily batch. Publishes what you cleared, inside your ceiling.
One decision a day rather than a stream of notifications. The queue is the whole interface: approve, kill, or edit the line. Anything you don't touch stays unpublished — silence is not consent.
Autopilot
Once you have stopped disagreeing with it
Ships inside the ceiling and the brand rules, and reports after.
This is the mode the product is for, and it is deliberately not the default. You should arrive here because you read four weeks of its decisions and kept agreeing, not because it was pre-selected during setup.
In every mode
What doesn't change
The dial decides who presses publish. It does not loosen any of these, including on autopilot.
The spend ceiling
A daily limit you set. Not a target it aims at and not a rule it follows — a number it has no path to exceed, in every mode including autopilot.
Brand rules
Words it will never use, claims it cannot make, colours it will not leave. Checked on each asset before it reaches your queue, rather than caught after something is live.
Reversibility
Every post can be pulled and every campaign paused from the notification that announced it. Revoking access on the platform's side stops it the same minute, without asking us.
The log
What it did, what it spent, and the reasoning it had at the time. This matters most in autopilot, which is exactly when you are not watching.
Practice
When to move the dial
The signal is boredom. When you have opened the morning queue five days running and approved everything without editing a line, you are no longer adding judgement — you are adding latency. That is the moment to move up one notch, not three.
The signal to move back down is the opposite: you find yourself rewriting hooks, or you have changed what the product is and the old brief no longer describes it. Step back to approve-first until the two of you agree again.
Related: your first thirty days and what it can't do.
Questions
Common questions
How much autonomy should I give an AI cofounder?
Start with draft-only for a week, move to approve-first once the output stops surprising you, and consider autopilot after about a month of agreeing with its decisions. The ceiling and brand rules hold in all three, so the dial only changes who presses publish.
Can I move back to a stricter mode?
Yes, in one setting, at any time. It is a dial rather than a one-way door — most people step back down when they change positioning and then move up again.
What is the risk of autopilot?
That it publishes something you would not have, inside your rules and inside your budget. That is a real risk and the reason it isn't the default. What it cannot do is overspend or make a claim your brand rules forbid.
Does more autonomy mean better results?
It means faster iteration, which usually means better results — but only once its judgement matches yours. Turning it up before that just gets you to the wrong place quicker.
Nothing to lose
Founding cohort seats include the first month and a written teardown of your current funnel.
Claim a seat100 seats. No card. We tell you on day 14 whether it worked.