A columnist on e27 wrote this week, in a piece published on 2 November 2026, that every meeting across the businesses they work with is recorded. An AI system turns each one into four lists: decisions, risks, blockers and next priorities. It takes minutes, and the writer relies on it. The headline carries what they noticed after months of reading those lists back to back. The AI prepared every decision, and the ceiling on scale did not move.

I think that observation is right, and I think the reason is simple.

The lists are useful, and they are the easy part

Meeting AI is sold on time saved. Minutes written for you, actions captured, nothing forgotten. Is there a real need? Yes. Anyone who has sat in a Monday meeting and then lost the decisions by Wednesday knows the need.

So what actually changed? Preparation got cheap. The decision is named, the risk is logged, the blocker has an owner. All of that now costs minutes instead of an hour.

What did not change is what happens after the list. Every item still lands on a person. That person opens the system, makes the call, sends the message, takes the payment, chases the supplier. The ceiling on a business was always the number of items its people can act on in a week. A better list makes that number more visible. It does not make it bigger.

Where the ceiling really sits

Take a clinic with two receptionists. The meeting AI records, every Monday, that no-shows are a blocker and that a deposit policy was decided. Next Monday it records the same two items. The decision was prepared in minutes. Carrying it out means someone sends each reminder, collects each deposit, and reschedules each cancellation by hand. The list is tidy. The receptionists are still the ceiling.

An online shop is the same. “Answer where’s-my-order faster” shows up as a priority week after week, because the answering is done by a person reading the order screen and typing the reply. A marketing team gets “follow up the leads from the webinar” as a next priority, and the follow-up is still a human working down a spreadsheet.

In each case the AI automated the chat about the work. The work itself is untouched.

Scale moves when the workflow runs, not when it is summarised

My position has been the same for a while. Automate the workflow, not the chat. The value is a process that runs end to end, and automation earns its keep when it checks, changes, books, charges or escalates inside the conversation, without waiting for a person to clear a list.

The clinic again. The reminder goes out on its own. The deposit is collected in the same chat. A cancellation is rebooked against the live calendar. The rules were set once, in the clinic’s own words: how far ahead to remind, how much deposit to take, which cases go to a person. Those limits on money, promises and dates sit in the system, not in a prompt. Rules where wrong is costly, AI where judgement is needed.

Next Monday, “no-shows” is no longer a blocker on the list. What remains are the exceptions: the patient who wants a refund above the limit, the booking that needs a doctor’s call. People decide those. That is a different job from clearing a list, and it is the job that scales.

This is the shape I build AeroChat around. You describe the workflow you want automated in your own words, the system translates it into instructions and runs it, and because it connects to your own data systems it can perform actions, not just answer questions. For one bakery merchant it handles 87 to 94% of conversations with no person, through festive peaks. That owner’s Monday list is shorter because the work ran during the week, not because it was summarised better.

I rely on preparation tools myself. I use Claude for almost everything and dictate to it with Wispr Flow. They cut the time it takes me to get to a decision. The thing that changed my capacity was the work that runs without me, inside rules I set once.

What a business should do with its meeting lists

  1. Take the last four weeks of AI-prepared lists and mark every item that appears more than once. A recurring item is a workflow wearing the costume of a decision.
  2. For each one, write down in your own words what should happen from start to finish, and the limits: how much money it can commit, what it can promise, which dates it can move. Name who handles the exceptions.
  3. Automate those workflows end to end, then read the next month’s lists. If the item disappears and only its exceptions remain, the ceiling moved. If the item comes back, the workflow is not running end to end yet.
  4. Keep the meeting AI. It is cheap and harmless, and the lists make the recurring items easy to spot. Just do not count it as scale.

Source: e27: AI prepared every decision. It did not raise the ceiling on scale