Overview
Work with AI and one person can handle three times the load. People say it all the time, and it is half true.
Your hands do triple. Decks, code, video edits: they come back in a fraction of the time they used to.
One thing does not grow. Your head. Take on more than it can hold, and the failure is decided at the moment you accept the work.
ANDOOR runs its projects with one founder and a set of AI partners. By AI partners we mean AI we hand work to, split by role. What follows was counted on that floor.
It breaks before and after the making
The making is not what breaks first. What comes before and after it is.
Take on three times the projects and the making itself holds up for a while. Your AI partners write, build and fix.
Something else drops first. Replies slip by a day. You start missing the morning stand-up. A premise agreed in last week’s meeting gets swapped with another project’s premise, and the work carries on built on the wrong one.
All of this happens before anything is handed to the AI.
Who gets a reply. Which meeting to attend. Which premise to build on. You decide these, and the number of decisions grows with every project you accept. Faster hands do not shrink it.
The end state: you ship well-made work, on the wrong premise, late.
Work built on the wrong premise is not what the client wanted, however carefully it was made. Each slip can be fixed the next day. But deciding whether to fix it takes the same head, and that decision joins the same queue. What the client sees is slow replies, meetings that do not connect, and rework. The quality of the work only gets seen behind all that.
What does “capacity” actually mean?
Call it capacity and you can never check whether you have it. Split it into three.
How many contexts you can hold at once
Every project has its own state: who decided what, where it is stuck, who moves next. How many of these can you keep in your head without mixing them up?
How long it takes to come back after switching
You leave the meeting for project A and open the deck for project B. How long until you remember B well enough to judge at full speed? More projects mean more switches, and the sum of those recovery times takes up real space in the day.
How many calls only you can make
Ship this wording or not. Send this estimate or not. Return this revision to the client or not. AI can lay out the options. The person whose name goes on the work makes the call. How many of those can you make in a day without the quality slipping?
AI does not grow these three. It fills them sooner
The faster your hands, the sooner your turn to decide comes around. Capacity stays the same and fills up faster.
A deck used to take half a day. For that half day, no new decision arrived. With AI, the deck is back in an hour, and the moment it lands, it is your turn to say whether it is good enough.
We counted October 21. Five meetings, spread across four projects. The same day, our AI partners sent up work from six projects, including ones with no meeting that day: a 20-row question sheet, a 9-slide meeting deck, three rounds of edits to a demo video, a set of screen designs, a 9-slide draft proposal, and a written response to a security advisory.
The hands can turn that much in a day. The head has to decide, for every item, whether it ships or gets reworked.
At ANDOOR, a separate AI grades the work the making AI produces. Twice that day, the grader sent something back three times without passing it, and the only one left to decide was the founder.
That day, something went wrong in a client’s shared sheet.
We had an AI partner add the 20 questions. A person on the client side was writing in the same sheet at the same time. Our instruction fixed the target row once and had the rows entered in one go, so we wrote over a row the client had just filled in.
The client re-entered their row, and we apologized. We bowed our heads over something that had nothing to do with the questions themselves.
The AI’s hands did not make the mistake. When we wrote the instruction, we did not have in mind that several people were editing the sheet at once. When capacity fills up, premises like that are the first thing to fall out.
October 7 gave a clearer number.
For each of our 22 projects, we had an AI partner put on one page how far it had come, where it was stuck, and who it was waiting on. Twenty-one projects had something stuck. Of those 21, seven were waiting on the founder. An unsigned document. A call on whether a site update could go out. A naming decision for a public page. Each would have taken a minute to answer if anyone had asked.
The hands were not waiting. The decisions were, and there was only one person who could make them.
Where the capacity came from
Capacity is not talent. It is what built up because there was no one to hand things to.
For 12 years, the work was done alone. Setting up a server while making a design, sending an invoice the evening after writing a proposal. No project premise could be handed to anyone, so all of them had to be carried alone.
With no one to consult, even the uncertain calls were made alone, and their consequences were owned alone. Twelve years of that, and it finally became possible to hold several projects in one head.
Even with all that, there is a ceiling.
The overwritten row on October 21 and the seven stalls on October 7 both happened on top of those twelve years. Now that AI has tripled the hands, the ceiling arrives sooner than it used to. Having capacity and being able to take on everything are different things.
Who gets to take on three times the work?
Only people with three times the head should take on three times the work. And even they only up to their ceiling.
Taking on work without knowing your ceiling is the most dangerous move. Measure first.
Count the decisions you made today, then subtract the ones you later reversed and the ones the client sent back. What remains is how many calls you can make in a day without the quality slipping. Then count how many calls one project asks of you per day. Divide the first by the second, and that is how many projects you can hold right now.
If it falls short, spend the extra hands on quality before taking on more. Make three options instead of one and compare them. Review three times before sending instead of once. Fix the problem before the client notices it.
If you want more volume, cut decisions, not hands.
Put into words the standards you decide by, and hand them to your AI partners. How to handle inconsistent wording, what to check in a review, how far you can fix without asking the client. Every decision you can put into words is one fewer call you have to make yourself. That much more work becomes possible.
ANDOOR can hold 22 projects because it has put many of its decisions into words and handed them to its AI partners. Even so, seven were stuck waiting on the founder.