How to think about knowledge work in the AI era
A team shipped more in one quarter than they had in the previous three. AI coding tools made them fast. Demos every week. Features flying out the door. Six months later, revenue had not moved. Customers were not happier. The team was exhausted and confused.
They were doing more work than ever. The work just did not matter.
This pattern is everywhere right now. AI has made us faster at producing things. It has done nothing to make us better at choosing what to produce or checking whether it worked. To see why, you need a simple model of how knowledge actually flows.
The loop
All knowledge work moves through six stages.
- Intent. Why are you doing this? The problem you believe is worth solving.
- Input. The raw material. Research, data, requirements, context.
- Process. The work itself. Writing, coding, designing, and analyzing.
- Output. The thing you produce. The feature, the report, the campaign.
- Outcome. What changes for someone because of your output? A customer saves time. A team makes a better decision.
- Impact. The lasting effect. Revenue grows. Trust compounds. The business moves.
Then the loop closes. What you learn from impact should reshape your intent. That feedback is the whole point. Without it, you are guessing forever.
I call this the Intent-to-Impact Loop. The two ends are where the game is now being played.
AI is hollowing out the middle
Picture a barbell. Heavy on both ends. Thin in the middle.
That is what AI is doing to this loop.
Gathering input used to take days of research. Now, a model summarizes a hundred sources in minutes. The process used to be the expensive part. A feature that took a team two weeks now takes one engineer an afternoon. Output used to be scarce. Now, anyone can generate a report, a design, or a working prototype before lunch.
The middle of the loop is becoming commoditized. Cheap, fast, and available to everyone. When something becomes available to everyone, it stops being an advantage.
The uncomfortable part is that most professionals built their careers on the middle parts. Their skill was the process. Their pride was the output. AI is compressing the exact stages where they live. Working harder at those stages now means automating yourself with extra steps.
Value is migrating to the ends
If the middle is cheap, where does value go? To the two stages AI cannot own. And both stages demand the same two human traits working together.
Intent demands taste and agency. Taste comes first. A model can do almost anything you ask. It cannot decide what is worth asking. Choosing the right problem, saying no to ninety-nine good ideas to pursue one great one. That judgment is taste. Then the agency takes over. A well-chosen problem is worthless until someone commits to it and starts. Plenty of people can spot a good problem. Far fewer take responsibility for it and act.
Impact demands agency and taste. Agency comes first here. Closing the loop takes effort that nobody assigns you. You have to go back, dig into what happened, and face the answer even when it is uncomfortable. Most people skip this because nothing forces them to do it. Then taste does the judging. Did this actually change anything? Was the change worth the cost? Is this good or merely done? Models can generate options and reports all day. They cannot tell you what matters to your customer, your business, or your life.
Taste to choose, agency to start. Agency to look, taste to judge. Both traits at both ends. The barbell again. These two traits were always valuable. AI just raised their price.
Most people run the loop open
Here is the second trap, and it is older than AI.
Most knowledge workers run the loop open. They form an intent, do the work, ship the output, and move to the next thing. They never wait to see the outcome. They never measure the impact. They never feed what they learned back into their next intent.
An open loop feels productive because shipping feels like progress. But shipping is only the fourth stage of six. A closed loop is how you compound. Every cycle teaches you something that makes your next intent sharper. Ten closed loops make you wise. A hundred open loops just make you tired.
AI makes this trap worse. When output is cheap, the temptation is to ship more and reflect less. The people who resist that temptation will pull away from everyone else.
How to work differently
Here is what this looks like in a normal week.
- Spend more time on intent. Before starting anything, write one sentence. What should be different in the world if this works? If you cannot write that sentence, you do not have an intent.
- Delegate the middle aggressively. Let AI gather input, draft the work, and produce the output. Your job in the middle shrinks to direction and review. Fighting this is like a calculator-era accountant insisting on long division.
- Schedule the loop closing. Put it on the calendar. Two weeks after shipping anything significant, spend thirty minutes asking what changed. Did the outcome happen? Is there a trail toward impact? What does this teach me about my next intent?
- Judge ruthlessly. When AI hands you ten options in ten minutes, your value is picking the one that matters and killing the other nine. Practice saying “this is fine, but it does not matter.” That sentence is taste in action.
The question to carry
The professionals who struggle in the next decade will be the ones who keep competing in the middle of the loop, racing machines at the machines’ own game.
The ones who thrive will own the ends. They will choose problems with care, let the machines run the middle, verify what changed, and feed every lesson back into a sharper intent.
So here is the question to ask about your own week. How much of your time goes to the middle of the loop, and how much goes to the ends?
Be honest. Then start moving toward the barbell.