AI's Bigger Win Is Subtraction, Not Creation

The first time AI genuinely changed how I think about work, I was not writing code.

I was staring at years of accumulated files in my personal Google Drive. Receipts, photos, half-finished projects, documents I had stopped using. Dozens of folders, organized loosely, full of small decisions I had been deferring for months. The problem was never finding the time. The real problem was the cognitive overhead: hundreds of small judgments about where things belong, made sequentially, for hours. Each one too small to be hard, but together exhausting enough that I would close the laptop and tell myself I would get to it next weekend.

I described what I wanted to AI. How I would organize it if I actually sat down with a clear head and a full afternoon. Done in under two minutes.

That moment did not feel like productivity. It felt like a category had been removed from my life.

The category most people don’t see

Most discussion of AI is about what it can produce. Generate this slide deck. Draft this email. Summarize this report. The demos are flashy because creation is visible. You can hold up the output and point at it.

The bigger value sits one layer beneath. It is not in what AI creates for you. It is in what AI lets you stop doing.

Asana’s 2023 Anatomy of Work Index surveyed more than 10,000 knowledge workers globally. The headline number is the one worth sitting with: 60 percent of the average workday goes to “work about work.” Searching for information. Switching between apps. Chasing status. Communicating about work instead of doing it. Reformatting things for different audiences. Triaging inboxes. The connective tissue of getting things done, not the things themselves.

A 40-hour week is, on average, 16 hours of actual work and 24 hours of overhead.

McKinsey Global Institute’s June 2023 report on generative AI puts a complementary number on it: across the activities that consume employee time today, 60 to 70 percent is now mechanizable. Asana measures where the time goes. McKinsey measures how much of it no longer needs a human.

The category hides inside busy days that feel productive. Each individual task feels necessary, because every task on the list is. The category becomes visible only when you offload it and watch what fills the space.

What belongs in “work about work”

It is easier to recognize by example than definition. Some that show up in nearly every knowledge worker’s week:

• Organizing files, photos, documents, downloads. The Drive moment above. Library work that has clear value at the end but no thinking at any single step.
• Meeting preparation. Reading three prep documents to extract the four questions you actually need to ask. The reading is necessary, but the reader-doing-the-reading is mostly format-conversion.
• Inbox triage. Sorting fifty emails into “respond now,” “respond later,” “archive,” “delete.” Pure classification with light judgment.
• Status writeups. Synthesizing the week of dashboards, threads, and decisions into a three-paragraph update for someone who was not in the room.
• Cross-team metric review. Twenty-seven dashboards across seven teams, each tracking a slightly different version of retention or engagement. The work-about-work is reading all twenty-seven and flagging the four or five that diverge meaningfully this quarter. Used to be a Friday afternoon. Now: feed the summaries in with rules about noise versus signal, get a draft list back in five minutes, and spend the remaining time on the diagnostic question the role actually exists to answer.

What ties these together is a particular shape. Each requires judgment, so a simple script cannot do it. Each is too small or too irregular to be worth hiring a person for. Each is too important to skip. So it lands on the desk of whoever is closest to the work, regardless of whether that person’s highest-value contribution is in this category.

For a generation of knowledge workers, this category has been a structural tax. Visible to anyone who tracks where their hours go, hard to delegate, hard to automate, hard to ignore.

Why this category resists traditional tools

The reason this work has been hard to remove until now is not that we lacked the will. We lacked tools with the right shape.

Spreadsheets handle data. Scripts handle deterministic steps. Project management software handles coordination. None of these handle judgment-driven, context-aware, slightly different every time. The Drive organization needed someone to know what “receipts” meant for my household versus a generic taxonomy. The meeting prep needed someone to know which of the four documents was actually relevant. The status writeup needed someone to know which audience required which detail.

You could automate parts. You could not automate the whole pattern, because the pattern required a thing that had been the exclusive province of humans: small, fast, context-aware decisions at scale.

AI’s distinguishing capability is exactly that. Not a single brilliant decision. A thousand competent ones, made quickly, against context the system is willing to take in.

This is why the demos about creation are misleading. They show a one-shot output: write me a poem, draft me a deck. The harder, more valuable use is volume judgment work. The kind that previously could not be delegated because it required taste, and could not be hired for because it required context.

What changes when you offload it

The instinct is to assume that offloading 60% of your work means producing 60% more output. That misreads what is being freed up.

A knowledge worker who reclaims ten hours a week of work-about-work does not produce 25% more deliverables. They produce roughly the same number with substantially more cognition available for the hard ones. The work most knowledge workers should be doing more of is bounded by cognitive load, not hours: hard decisions, contested analysis, original strategy, real conversation with stakeholders. You cannot run two hard meetings back-to-back at full quality. You can run six work-about-work hours back-to-back without noticing.

The math is not “10 hours back, 25% more output.” It is 10 hours back, and the remaining hours run cleaner.

The trap

Subtraction is harder to see than creation.

When AI writes a draft, you can point at it. When AI removes three hours of file organization from your week, the result is invisible: three hours that simply do not exist anymore. Your calendar does not record a thing that did not happen.

Two practical reframes:

• Audit last week by category, not by task. Group the hours by what kind of cognition each task required. Most of us discover that one or two categories ate most of the time without producing most of the value.
• Identify the category that is too small to hire for but too consistent to ignore. That is the first AI target.

The Drive moment again

The Drive itself was not the point. Two hours saved on a Saturday is not life-changing. What changed was the realization that an entire category of cognitively expensive, strategically worthless tasks had been silently consuming my time for years.

A year later, the most useful question about my work week is not what AI helped me create. It is what AI helped me stop doing. The list of things I do not do anymore is the real productivity gain. The deliverables look about the same. The work behind them runs lighter.

What was the first thing AI took off your list that you had not realized was optional?