The Plan Before You Leave the Room

The technology has been here for over three years. Everyone has it in Teams. And still: strategy days, leadership meetings, planning sessions. You sit there afterwards building the plan by hand in PowerPoint, from memory, exactly like five years ago.

It makes me angry, honestly. Not because the technology is hard. Because the simple thing doesn't get done.

You get stuck on the wrong question

Every time someone tries recording a meeting, the conversation is the same. The transcription isn't quite right. There's an ambiguous phrasing. Which model is actually best. After three years of this I know all of that is secondary.

What matters is getting your own content in. The conversation, together with the notes, with the agenda, with the plan you already sketched. It's the context that does everything. The prompt is the small part. And most of the energy goes into exactly what matters least.

The simple thing

Throw the conversation, the notes and the agenda into the same query. Claude, ChatGPT, it matters less -- but a paid service. The free version gives you a truncated summary. And when you're working on a strategy, truncated is the same as worthless. You need enough data out.

Then a human sits in the loop. The first answer isn't the truth. You iterate two, three times. You catch the nuance that fell away, the person who didn't get the right task. It's a process and you steer it. But the result isn't remotely like a plan you wrote yourself afterwards, alone, at six in the evening.

Why it must not be automated

The thought always shows up here, and it shows up fast: can't we just automate this? Build a flow that records, transcribes and produces the plan by itself?

No. And it's worth being clear about why, because it isn't a technical objection.

There's no intelligence in the model weighing your specific situation. It's a prediction of what others have phrased before. It doesn't know which of your departments is already overloaded, that the person who got the main responsibility is leaving in a month, that last year's plan failed on exactly the point you're about to repeat. It can't know what you didn't write in. And it doesn't care what the conclusion belongs to.

That's why the human in the loop isn't a quality check you bolt on at the end. It IS the work. It's in the iteration -- when you read the first draft and say no, that's wrong, that isn't what we meant -- that the plan actually becomes yours. Automate that step away and you've automated away the value and kept the transcription, which was the part that mattered least.

A strategy plan you spent a full day on is too important to hand to a third-party flow that nobody owns. The point isn't to save away the meeting. The point is to move the meeting discussion into an action plan. And that move is a human one.

Never ask it yes or no either. A prediction system should structure your work, not make your decision.

Do it before you go home

The best moment is the final session. You've had your day, you've discussed, the whiteboard is full. Before you break: press the button, send in the day, put the result on the screen. Share it. Reflect together. Adjust right there while everyone is still present and remembers what was said.

Then nobody leaves with the feeling of a good meeting and nothing more. Everyone leaves with the day's action plan. Tomorrow you start working instead of summarising.

The difference between the two is one button press and a human willing to iterate. That so few take that difference is the only hard thing in the whole matter.

See also: Give me the insight, not the decision (series 34) and It's the I in AI That Does It (series 39).