How AI preps me for every meeting in the time it takes to make a coffee

There is a particular kind of dread that hits about four minutes before a client call. You know the one. The meeting is in the calendar, you accepted it a fortnight ago, and now it is nearly on you and you cannot for the life of you remember where you left things. What did they ask for last time? Did that email ever get answered? Is there a task sitting open that they are about to ask about?

So you scramble. You skim the last email thread, you half-remember the previous call, you open the project to check what is outstanding, and you join the call slightly on the back foot, hoping nothing comes up that you have forgotten.

For a small business that is not a disaster once. It is a disaster as a habit. Multiply a few minutes of scramble and a bit of looking unprepared across every client, every week, and you have a real cost, in time and in how buttoned-up you seem to the people paying you.

So I built the prep once, properly, as something an AI does for me. Now I say "meeting prep for [client]" and about thirty seconds later I have everything I need, pulled fresh from the actual sources, laid out and ready to read.

What it actually does

Say I have a catch-up tomorrow with a client. I ask for the prep, and the AI goes and gathers, in one pass:

• Every open task for that client, straight from the system I run all client work through. Not my memory of what is outstanding, the live list.
• A summary of our last few meetings, pulled from the notetaker that sits in every call and transcribes it. So "what did we agree last time" is answered from what was actually said, not what I think I remember.
• The recent email traffic with that client, the last month of it, so anything still hanging in the inbox surfaces before the call, not during it.

Then it does the part that actually saves the day: it reads all of that and hands me back a single structured brief. Open work, listed. What has happened since we last spoke, summarised. A suggested agenda, built from the tasks that need a decision and the emails that need a verbal follow-up. And a short list of questions I should be asking them, usually the things currently blocked waiting on their side.

It even reminds me of the meeting habits I am personally working on, a couple of lines at the top about how I want to show up, because prep is not only about the content, it is about how you run the room.

I read down the page, tweak anything that needs tweaking, and I am ready. The whole thing takes about as long as making a coffee, and I walk in knowing exactly where we are.

An example of the brief the skill produces, fictional client, real structure. The open work pulled live, a summary of where we left things last time, an agenda built from what actually needs a decision, and the two questions I need to get answered on the call. All of it gathered before I sat down.

Why it beats doing it by hand

The obvious answer is speed, but that is not really it. I could skim three sources by hand in a few minutes if I had to.

The real win is that it never forgets a source. When I prep by hand under time pressure, I check the two things that are top of mind and miss the third. The email I forgot about is exactly the one that comes up. The AI checks all three every single time, in the same order, whether I am fresh on a Monday or frazzled on a Friday. It is not smarter than me, it is just never in a rush and never distracted.

The second win is that the prep does not vanish. It gets written into the meeting record itself, so the brief lives with the meeting. Next time I open that meeting, the prep I did is right there, and it becomes part of the history of the account. Nothing is done twice.

How it works

Under the bonnet this is a skill, which for me just means a plain instruction file that tells the AI, step by step, what to fetch, how to think about it, and what to produce. There is no big platform here. If I want to change how prep works, I edit a document.

The skill does two things. First, it makes sure a meeting record exists (creating one if not, linked to the client and, if relevant, the specific project). Then it runs the prep.

It fetches from three sources at once. Tasks, meeting transcripts, and email all get pulled in parallel rather than one after another, so the wait is the slowest single source, not the sum of all three. Three quick gathers happening together, not a queue.

It reads summaries, not everything. This is the decision that matters most, and it is a deliberate restraint. When it pulls past meetings, it takes the summary of each, not the full word-for-word transcript. A full transcript of three hour-long calls is a wall of text that costs more to process and, more importantly, buries the point. The summary is what I need to walk in informed. Same with email: it reads the subject and the snippet, not the entire thread. You pull the least you need to be useful, and no more. Restraint is a feature.

The output is shaped to where it lives. The brief gets written into a notes field in the meeting record, and that field renders a limited subset of formatting. It shows headings and bullet points, but it will not render a table, a table just comes out as raw text and looks broken. So the whole brief is built from headings and one-line bullets on purpose. It is a small thing, but it is the kind of small thing that separates a tool that works from one that looks nearly right and annoys you every time. You build the output to fit the surface it lands on.

Alongside the notes, the skill also produces a properly branded one-page version of the brief, saved with the client's files, for when I want the prep as a clean document rather than a field in a database.

Nothing happens without me. The skill gathers and drafts, then stops and shows me the brief before it writes anything anywhere. I can accept it, amend it, or add an agenda item that only I know about. Only then does it save. The AI does the legwork, the low-judgement work of fetching and sorting and first-drafting. I keep the judgement.

The honest catch is the same one that applies to any tool like this: the prep is only as good as what is actually recorded. If a task never got logged, it will not appear in the open work. If a meeting had no notetaker in it, there is nothing to summarise. The brief reflects your systems faithfully, including the gaps in them. That is not really a flaw in the build, it is a nudge to keep the underlying records honest, because now they pay you back directly every time you meet a client.

The loop: prepared questions become answered ones

Here is where it stops being a nice-to-have and starts earning its keep. The prep does not end when the meeting starts. It sets up the next step.

Look back at that brief. It listed a couple of specific questions I needed answered, the vector logo, the rule for duplicate learners. Those are not just reminders for me, they are a checklist the whole loop is built around. I go into the call knowing exactly what I need to come out with.

And the meeting itself is being recorded and transcribed the whole time, by the same notetaker whose summaries fed the prep. So the moment the call ends, there is a full record of what was actually said, and sitting right beside it, the exact questions I walked in wanting answered.

That is a perfect setup for the AI to do the next job. It reads the transcript against the prepared questions and pulls out the answers: the logo is coming Thursday, duplicates should keep the most recent record. It surfaces anything that was agreed, any new task that came up, any date that got committed to. What was a list of open questions before the call becomes a list of decisions after it, extracted from the real conversation rather than from my scribbled notes.

From there the answers flow into whatever the meeting was for:

• Next steps and tasks. The commitments made on the call, mine and theirs, get turned into actual tasks in the system, so nothing agreed verbally quietly evaporates.
• A scope of works. If the meeting was a scoping call, the answers to those prepared questions are precisely the raw material a scope document needs. The AI drafts the first version from the transcript, and it is grounded in what the client actually said they wanted, not my interpretation a day later.
• A proposal or a quote. Same principle. The decisions pulled from the call feed straight into the numbers and the wording, so the thing I send back reflects the conversation we just had, tightly.

So the shape of the whole thing is a loop, not a one-off. The prep gathers what I need to walk in ready. I walk in and get the answers to the exact questions I prepared. The transcript captures those answers. The AI reads them back out and turns them into the tasks, the scope, or the proposal that the meeting existed to produce. Each stage hands the next one something concrete, and the questions I set at the start are the thread running all the way through.

That is the part that changes how meetings feel. A meeting is no longer a thing you prepare for, sit through, and then have to write up from memory. It is one step in a chain that mostly assembles itself, as long as you know the questions you are trying to answer going in.

What you can take from this

You do not need my exact setup to use the idea. The pattern travels to any business that has meetings and hates prepping for them.

• Prep is a gathering problem, not a thinking problem. The hard part of walking in prepared is not clever analysis, it is reliably pulling the same handful of sources every time without forgetting one. That is exactly the kind of boring, repeatable job to hand to a machine.
• Pull from the live source, not your memory. The value is that the tasks, the notes, and the emails are the real current state, gathered fresh, not your recollection of them from a fortnight ago.
• Take summaries, not everything. More text is not more useful. The skill that reads three summaries beats the one that dumps three transcripts on you, every time. Decide what "just enough to be useful" looks like and stop there.
• Let it draft, keep the decisions. Have the machine assemble the brief and propose the agenda, then read it, correct it, and own what you walk in with. Approving is faster than doing, and you stay in control.
• Set the questions going in, harvest the answers coming out. The questions you prepare are the same thread that turns a transcript into next steps, a scope, or a proposal afterwards. Prep and follow-up are two ends of one loop, not separate chores.

Close

The meetings themselves were never the problem. The scramble in the four minutes before them was, and the slightly unprepared feeling that came with it. Fixing that did not take a grand system, it took one instruction file that gathers three things I was already checking by hand, and checks them more reliably than I ever did.

If there is a moment in your week you dread out of pure unpreparedness, that is the one to look at first. The odds are the information you need already exists, scattered across a few tools, waiting for something to go and fetch it before you need it rather than while you do.