When Did We Start Rewarding Waste Instead of Real Productivity?

For most of my career, productivity has been pretty easy to understand. What did you produce, how long did it take, what did it cost, and was the result worth the investment?

Then AI came along, and somewhere in the excitement I think we started measuring something else.

We started talking about millions of tokens consumed, enormous context windows, hundreds of agent actions, and increasingly complicated workflows as if the amount of AI being used were itself evidence of productivity.

I am not convinced it is. In fact, I think we may be rewarding waste.

If I hired someone to build a house and it took them ten years to build something another contractor could build to the same standard in three months, I would not congratulate the first contractor for all the additional labor they used. I would want to know what went wrong.

Why should AI be measured differently?

I use AI constantly, and I have also wasted an enormous amount of time using it. That experience is actually what made me start thinking about this.

Not long ago, I was creating an image. I had one AI-generated version where the image itself was exactly right. I had another version where the text was exactly right. All I needed was the right text on the right image.

It sounds simple.

The AI apparently disagreed.

I supplied the examples. I explained what I wanted. I prompted it again. Then again. It would fix one part and change another. It would get close and then somehow wander away from what had already been correct.

After five or ten minutes, I realized I was being ridiculous.

I opened Photoshop, cut the pieces I needed, put them together, and finished the job in about five minutes.

I could have kept going with AI. I could have burned another twenty minutes and thousands more tokens trying to convince it to do something I already knew how to accomplish.

And that is where I think our current way of talking about AI productivity gets strange.

If using more tokens is considered evidence that I am becoming a more advanced AI user, then continuing to fight the machine would have looked better on paper.

But I would have been less productive.

The objective was not to use AI. The objective was to finish the image.

That distinction matters.

I think the best AI users of the future will not necessarily be the people who use the most AI. They will be the people who consistently get the best results using the least amount of AI necessary to accomplish the job.

That does not mean using the fewest tokens at all costs. Some problems deserve enormous amounts of computation. If deeper research, additional reasoning, more testing, or a larger context produces a materially better result, spend the tokens.

But if two people create an equally good result and one requires two million tokens, fifty attempts, an hour of employee time and several corrections while the other reaches the result with three prompts and a ten-minute manual adjustment, the second person did the better job.

That is not some new theory of artificial intelligence.

That is productivity.

More importantly, that is ROI.

Return on investment has always been one of the most reliable ways to understand whether a business, department, employee, campaign or piece of equipment is actually creating value. What did we put into it, and what did we get back?

AI does not change that formula.

Tokens cost money. Employee time costs money. Prompts cost time. Failed generations cost time. Reviewing AI output costs time. Correcting hallucinations costs time. Rework costs money. So does accepting mediocre work because everyone got tired of asking the machine to fix it.

The value on the other side of that equation is the finished result.

That brings us to something else I think businesses may be getting wrong about AI: expertise.

There is an assumption that because AI allows more people to perform specialized tasks, experienced people will become less valuable.

I suspect the opposite may happen in many fields.

The person who does not know Photoshop might reasonably spend an hour trying to get AI to produce my image. Their alternative might also take an hour because they would need to learn how to create it manually.

For that person, fifty attempts with AI may still represent the most efficient option available.

But my economics are different.

If I can finish the problem manually in five or ten minutes, then AI has a much shorter window in which it has to prove useful.

My expertise changes the break-even point.

That is an important distinction because an expert does not merely know how to use the tool. An expert has judgment.

They know what the finished product is supposed to look like. They can recognize when something is almost right and when it is fundamentally wrong. They understand what AI does well and where it tends to struggle. And when the machine stops being the fastest way to reach the target, they have another way to get there.

That may change the role of experienced professionals considerably.

In the past, we often thought of the expert as the person who painstakingly built the thing from beginning to end.

With AI, the expert can become something closer to a sniper.

They understand the target. They know where to aim. They use AI aggressively when it gives them leverage. They may accomplish in an afternoon what previously took several days.

But when AI misses, they do not have to stand there firing indefinitely.

They know when to switch tools.

There is an old story that has been told in various forms about an expert being called into a facility after a complicated piece of machinery fails. Nobody at the company can get it working. The expert walks around, studies the equipment, makes a small adjustment and the machine starts running again.

Then he sends a very large bill.

The owner objects. How can you possibly charge this much? You were only here for a few minutes.

The answer, of course, is that the customer was not paying for those few minutes. They were paying for the decades of experience that allowed the expert to know what to do during those few minutes.

AI may make that kind of expertise more valuable, not less.

Perhaps businesses will eventually stop measuring talented people primarily by how many hours they spend producing something and become more comfortable paying them extremely well to solve important problems quickly.

You already see this with great lawyers, surgeons, consultants and other specialists. You are not paying top dollar because you want the problem to take longer. You are paying because you want somebody who has seen the problem before, understands what matters and knows how to get to the right answer.

Efficiency is part of what you are buying.

AI simply gives those people additional leverage.

That is why I think we should be careful about the metrics we establish now. If companies begin rewarding employees because they consumed more tokens, generated more prompts, or spent more time interacting with AI, we risk recreating one of the oldest management mistakes in a very expensive new form.

We will be measuring activity instead of output.

The best AI employee may eventually look surprisingly quiet on the dashboard. They might use fewer prompts. Fewer tokens. Fewer agents. They may occasionally close the AI tool entirely and finish the job themselves.

But if they consistently produce excellent work faster, with fewer mistakes and at lower total cost, that is the person creating value.

We do not need a new definition of productivity because artificial intelligence arrived.

We need to remember the old one.

What did we create?

What did it cost us?

What was the return?

This philosophy has also influenced the way we are building Potomac Lux.

Owning a home requires expertise across dozens of areas, and no homeowner can reasonably be expected to know all of them. One person may understand HVAC systems. Another may know construction. Another may barely know where the water shutoff is.

That should not determine whether someone can take good care of their home.

The point of applying AI to Potomac Lux is not to make homeowners become better AI users. It is to reduce the amount of expertise they personally need to carry around in their heads.

Sometimes technology can recognize the problem. Sometimes it can organize the information or surface the right question. Sometimes an experienced contractor needs to make the call. And sometimes the smartest thing the technology can do is step aside and let a professional handle something they can solve faster.

We want Potomac Lux to be that sharp set of eyes in the background, helping the homeowner get to the right answer without making them manage the machinery behind it.

Because ultimately, nobody should care how many tokens it took.

They should care that the problem got solved.

That is the metric.

Do not show me how much AI you used. Show me the return.