Audit your attention, not your time

Founders know their numbers. Cash in, cash out, runway, the margin on every product. Most of us could not say, with the same precision, where our own attention went last week. For a company of one or two people that is the stranger of the two gaps, because attention is now the scarcer resource.

Here is why. Ten years ago I ran a flower import business in Ukraine, and every piece of software it needed, from the websites to the CRM to the mobile apps, was built by a development team I paid. Today I am co-founder and CTO of a biotech startup, and I have written its entire IT infrastructure myself with AI tools, without a single developer. Code has become as cheap as electricity. The work that used to fill a small team's week now fits into the hours one person can point a machine at.

When execution stops being the bottleneck, something else becomes it. In my case, and I suspect in most founders' cases, it is the choice of what to point the machine at. The agents will do whatever I give them. The only thing they cannot do is decide what deserves doing. That decision is made, hour by hour, by wherever my attention happens to land.

So I started auditing it. Once a week an AI agent reads through what I actually worked on, and writes a short report on where my attention went. I set it up for a simple reason: I wanted to see how much of my focus went where, whether that was rational, and whether something needed correcting. Some weeks it does. Claude now ships a monthly version of the same idea to every user, which tells you the people who build these tools think the question is worth asking.

What the audit catches is rarely dramatic. It is the Eisenhower matrix, the old grid of important against urgent, playing out without anyone watching. The important and non-urgent tasks, the ones that only get done if you force your hands towards them, quietly starve. The urgent and unimportant ones, which arrive by email and feel like work, eat the week. Time-tracking tools have never fixed this, because they measure hours. They do not tell you that the hours went to the wrong quadrant.

The fix follows from the diagnosis. The best model I know for working with AI is a new junior employee: capable, fast, speaks every language, but not you. You do not tell a junior to build the business. You hand them the urgent-and-unimportant pile, the translating, the sorting, the first drafts, and you keep the thinking. A weekly attention audit is the list of what to hand over next week. If the report shows three hours on formatting a spreadsheet, that is three hours a junior should have had.

There is a trap on the other side, and the audit catches that too. Eliyahu Goldratt, in The Goal, has a plant manager boasting about his new machines. A friend asks three questions. Did you ship more? No. Did your inventory fall? No. Did you spend less on labour? No. Then you automated nothing. AI invites the same self-deception. A founder who "saved ten hours with AI" has saved nothing unless the hours landed somewhere: more shipped, something new started, a cost that disappeared. Saved hours are not saved money. An attention audit forces the question of where the hours actually went, and the honest answer is sometimes "into more email".

A few practical notes for anyone who wants to try this. Keep the report to three questions: where did my attention go, did it match what I said mattered, and what should be handed off next week. Let the model read what it can see, your calendar, your documents, the tasks you closed, rather than asking you to log anything by hand; a system that needs manual input will be abandoned by Thursday. And run it for at least a month before judging it. The first weeks mostly show how far a week drifts from the plan, which is uncomfortable and not yet useful. Patterns appear later.

I run my startup's technology alone, I run my own video channels, and I keep a small app on the side that gets about five per cent of my attention. None of that would be possible for one person without agents doing the hands-on work. But the agents do not make the week. I do, by deciding where to look. For a founder in 2026 that is the one input the machines cannot supply, and the one almost nobody measures.

‍

About the author: Khrystyna Didukh is co-founder and CTO of Avara Bio, a biotech startup selected among the top 9 at TechCrunch Disrupt Startup Battlefield, a former PwC consultant, an official Envoy of Estonia's e-Residency programme and a UNDP Woman Innovator. She trains corporate teams on AI adoption.