Sit through enough status meetings and you notice something odd. The meeting isn't where the work happens — it's where people describe work that happened somewhere else, and agree on work that will happen somewhere else later. Nothing actually gets made in the room. It's pure overhead, sitting on everyone's calendar every week, rarely questioned.

Now think about how your organisation measures the value of AI. Almost certainly, it counts tasks: hours of drafting saved, support tickets resolved without a human, reports that once took days and now take seconds. All of it real, all of it countable, and all of it missing the bigger point.

Because the largest thing AI does for a business has little to do with the work itself. It removes the work that surrounds the work: the briefing that no longer needs a meeting, the status update nobody has to write, the approval that stops sitting in an inbox for three days. That's the coordination tax, and it's where the real return has been hiding.

Where knowledge work actually goes

We tend not to look too hard at this number, so here's the figure. Asana's Anatomy of Work research into how people spend the working day has been consistent for years: knowledge workers lose around 60% of their time to what the company calls “work about work” — chasing updates, sitting in meetings that needn't exist, hunting for information, switching between tools. Six hours in every ten, spent not on the work itself but on the work that surrounds it.

And it isn't only Asana pointing at this. Rob Cross, writing in Harvard Business Review, has tracked the same drift for years: collaboration time creeping upward decade on decade, until at a lot of firms it leaves only a sliver of the week for the work people were actually hired to do.

I've written before about why this makes teams feel so expensive, and why the answer isn't to cut the people. What I want to do here is narrower: look at why, when AI takes a bite out of that 60%, almost nobody notices.

The savings you can't see

Task automation is easy to measure because it leaves a trace. You can point at a finished report and say that it took an hour and now takes a minute, then put the comparison on a slide.

Coordination removal is the opposite. Think about the Monday status call that everyone privately dreads — the one where eight people take turns describing what they did last week. Give that team an AI that holds a shared, current picture of the work, and within a month the call has shrunk to a five-minute glance at a summary. Nobody logs that. There's no line item for the forty minutes you stopped losing every Monday. The saving is real — often far larger than any single task saving — but it shows up as an absence, and absences don't appear on dashboards.

I can't pretend this is easy to measure. I'm not sure it can be — you can't audit a meeting that never took place, and I'm wary of anyone who claims a clean number for it.

But hard to count doesn't mean small. It usually means the opposite. The things we measure most carefully tend to be the things that were easiest to measure, not the things that mattered most.

This is why so many companies feel faintly let down by their AI return. They went looking for it where it was easiest to count, found a modest number, and concluded the technology had been oversold. They measured the visible saving and never weighed the invisible one, which was the whole reason to bother.

What AI actually removes

Strip away the abstraction and AI cuts coordination in three fairly concrete ways.

  1. Shared context removes the briefing loop. Most briefings exist because knowledge is trapped in one person's head and has to be transferred into another's before work can continue. Put that context — the project history, the customer's earlier emails, the last decision and the reason for it — into a system that everyone, and an AI, can read, and the briefing stops being a meeting. It becomes a question you ask at the moment you need the answer.
  2. Async decision support removes the alignment meeting. Alignment meetings exist to get a group to a shared answer. Much of what happens in them isn't genuine disagreement; it's people arriving with different information. When an AI can lay the options, the trade-offs and the relevant data in front of each person before they weigh in, the meeting gets shorter or stops being needed. The decision turns asynchronous.
  3. Automated handoffs remove the wait. The most expensive phrase in knowledge work is “waiting on someone.” A handoff is a pause: work stops, joins a queue, and waits for a human to pick it up. Automate the routine handoffs — the draft that moves itself to review, the data that gets pulled and formatted without anyone asking — and the queue drains. Work that took a week of mostly waiting takes an afternoon of mostly working.

Your AI dashboard has no row for any of this. It counts tasks. What I'm describing lives in the spaces between them, which is where the week actually goes.

The longship advantage

A small crew moves faster than a large one, and not because anyone is rowing harder. They've simply no coordination tax to pay. When everyone can see everything and decisions get made in the moment, alignment takes care of itself. It becomes the water you swim in. I've used the image of a Viking longship for this before: a small business at its best is a crew where everyone rows and nobody needs a meeting to find out where the ship is heading.

What AI does, at its most valuable, is let a larger group behave a little more like that small crew, not by hiring people whose job is to coordinate, but by removing the need for them. The longship's crew shared context because they could all see the same horizon. A larger team can now share context because they can all see the same system.

A large organisation, watching a faster rival, tends to measure the wrong thing — how hard its own people are rowing, how many tasks they closed. That misses the point entirely. The rival pulled ahead by carrying less drag, not by straining harder at the oars.

Stop counting tasks

None of this means task automation is worthless. It plainly matters. But it's only the visible tip of something much larger, and because it's visible it collects all the attention. The danger is that you optimise for the tip: you buy AI to draft faster and close tickets quicker, you measure precisely that, and you never go looking for the prize underneath.

So there's a better question than “what tasks can AI do for us?” Try asking “what coordination can we stop doing?” instead. Walk through a normal week and mark the meetings that exist only to move information from one head to another, the updates written purely so someone else knows what's going on, the approvals queuing in inboxes. That scaffolding around the work is exactly what AI can take down, and the leverage has been sitting there all along, in the meeting that never gets booked and the email that never gets sent.