AI Took the Paperwork So I Could Be in the Room

TL;DR: A month into running a twelve-month system build on site, I’ve handed meeting transcription, minutes, reports and reminders to AI that only drafts, and I check every fact against the source before anything goes out. The time I save goes to people work: agreeing who speaks for what, how decisions get made and how the team communicates across time zones. It also goes to the human parts of the job no tool can do, like building trust, reading the room and owning the decisions, because the AI took the paperwork so I could be in the room.

---

My first month on site as project manager of a twelve-month build was also the month I set up a way of working with AI, so my time goes to delivery instead of administration. None of it is clever. Most of it is plain habits. But a month in, I can say where my hours go, and very few of them go to paperwork.

What the AI does

Every meeting is recorded and turned into text on my own laptop. Nothing leaves the machine.

Our meetings switch between Indonesian and English, sometimes in the middle of a sentence. I tried the smaller speech models first. Only the largest one got people’s names right. It is slow, too: it takes about as long as the meeting did. I don’t mind.

One small fix made a big difference. I keep a list of project names and terms, and the transcriber uses it to spell them. Without it, the project names came out as near-misses.

From the transcript, the AI writes the minutes using my minutes template. From the same material it drafts:

• the weekly report
• agendas
• follow-up messages
• reminders

Each one uses my own template. The AI can also read my email and the project tracker. That means a draft starts from what was actually said and what is actually written down, not from a blank page.

The rule I don’t bend

The AI only ever drafts. I check everything and send everything myself.

I learned the second half of that rule early. Sometimes the AI states a fact from memory when it should read the source. A date that was right last week. A status that has since changed. It reads well and it is wrong.

So every fact in a draft gets checked against the email or the ticket before it goes out. It is the part of the job I won’t hand over.

Where the freed time went

Month one was mostly people work.

On paper, everyone is on the project. In practice, I needed to find out three things: who actually decides, who actually knows, and who is only copied in. Those are often three different people, and the org chart doesn’t tell you which is which.

Once I had a fair picture, we put a few things in writing.

A voice map. One person speaks for delivery and schedule. One speaks for architecture. One speaks for scope, money and data. When a question comes up, we already know whose answer counts.

A decision rule. A hard-to-reverse decision needs an explicit yes, and silence never approves one. Everything else goes ahead on my recommendation unless someone objects by a stated time.

One clock. The whole project runs on one time zone for dates and deadlines. No more “your Tuesday or mine?”

Written first, calls last. Inside the delivery team, everyone writes three lines of update a day instead of joining an internal stand-up. We only get on a call once writing has failed twice on the same item.

That last one wasn’t about saving meetings for its own sake. The team is spread across time zones, and our overlap with the client’s working hours came to about one usable hour a day. One hour is too precious to spend on status updates. It is also too short to make anything wait for it. So a reversible decision never waits for anyone. It goes ahead, and we fix it later if we need to.

The part AI can’t do

All of the above is setup. The work that actually moves the project is human, and no tool comes near it.

Building trust in the first weeks. People watch how a new project manager behaves before they believe anything they say. You earn it by turning up, doing what you said, and not overreaching.

Working out who really holds the whole picture. Every project has someone who knows how all the pieces fit. They are rarely the most senior person in the room. Finding them early saves weeks.

Asking a favour once. I asked a builder for an early-morning session once. They said yes. I won’t ask twice. A favour stays a favour only if you don’t turn it into a habit.

Knowing who does what in a demo. In a client demo, the person who built it drives the screen. The project manager runs the room: the time, the questions, the “let’s take that offline.” Mixing those two jobs up makes both worse.

Asking in writing, one item per line, each with a date. A paragraph of requests gets a paragraph of goodwill and nothing done. A list with dates gets answers, line by line.

Noticing what isn’t said. Sometimes a person answers every logistics question quickly but never commits to a number. That pattern is information. A transcript records the words. It won’t flag what’s missing.

Knowing what stays in conversation. Some things belong in a quiet word and nowhere else. They stay out of minutes, out of reports, and out of anything that gets passed around. Deciding where that line sits is judgment, and it’s mine to make.

Owning the decisions. The AI can draft the recommendation. It can’t stand behind it. When a call turns out wrong, the name on it is mine, and it should be.

What a month taught me

The paperwork didn’t disappear. Minutes still get written, reports still go out every week, reminders still land on time. I just stopped being the one who types them from scratch.

What I got back was time in the room. Time to watch who looks at whom before answering. Time for the hallway question that saves a week. Time to be the person people come to before a problem turns into an email thread.

That was the trade I wanted. The AI took the paperwork so I could be in the room.