AI doesn’t cut people. It multiplies tasks

The main promise of AI for business sounds simple: work gets automated, and you need fewer people. Every new tool is sold exactly this way — save time, save on staff.

I look at it from inside a company that builds a productivity product itself, and I see a different picture. AI really does make people more organised and more productive. But it’s in no hurry to reduce their number. Instead, it multiplies tasks.

And that’s where the question hides that every owner will ask sooner or later: what am I actually paying for?

More tasks isn’t more results

Today everyone in a company uses AI. But it’s very hard to see where exactly it lifts productivity.

People got a tool that produces a plan, a backlog, a list of ideas or ten versions of a text in a minute. And they started doing more tasks. Not showing more results — doing more tasks.

Creating a task is now easier than ever. Closing it is as hard as it always was. So the list grows, activity grows, and whether anything the business can count is growing remains an open question.

The stack gets pricier with every user

Take a typical modern setup: an AI agent that connects to various apps and files everything into Notion. It looks great. But look at how the cost of the productivity stack grows.

You pay for Notion. You pay for AI tools, all of which are fairly expensive. And you pay for every user.

Of course, all of this raises productivity. But enough that you could, say, start saving on people? A stack like this seems to make people more organised automatically, yet it doesn’t reduce their number. The company pays for the person and for everything that helps them — and both lines of spending go up.

The founder as a manual aggregator

There’s one more cost nobody puts in the budget: the founder’s time.

The state of the company is now scattered across LLM chats, Slack, email and a dozen other sources. Someone has to assemble it into one picture: what’s done, what’s decided, what’s on fire. In my observation, founders do this either in their heads or by hand through reports — and it takes from a third to half of their time.

I once listened to a podcast with an investor who put it very precisely. He doesn’t want to hire more people into his fund. He wants the people he has to automate everything with AI. But for that you need a system for managing AI agents — they need organising too, and right now that’s the problem.

AI didn’t remove management. It added one more layer to manage.

Measure results, not activity

On our team we tried to rate people’s productivity automatically from their Slack activity. The conclusion the model itself reached was honest: the system measures activity, not the actual result of the work.

A “quiet” day looks unproductive even if that day a result shipped to production. The system can’t tell apart a person being away, missing data, work done outside Slack, and no result at all.

It’s the same trap as with AI in general. We count what’s easy to count: messages, tasks, generated pages. That counts effort, not value. Something simply updated in a repo is already a commodity anyone has. Value appears when what gets updated delivers a real gain in productivity.

Where to really look for the effect of AI

If AI doesn’t cut people automatically, its effect shouldn’t be looked for in the headcount, but in the visibility of work.

A company gets a return on AI when it can see where AI genuinely lifts results and where it only adds tasks. That needs not one more source of activity, but a place where the work of people and their agents comes together into a clear picture: what’s done, what’s decided, what’s moving and what’s stuck.

Then the question “how many people do we need” turns into another one: “what are we spending the people we already have on.” And the answer can finally be seen rather than guessed.