Last month I wrote about a question my wife asked me. If AI has made me so much more productive, why am I busier than ever? I answered with the Jevons paradox. When fuel gets cheaper, people burn more of it, and total consumption rises. My fuel was my time, and my remedy was the one James Watt applied to his steam engine: a governor that throttles the machine before it runs away with itself. My resolution - I would cap my hours.
I failed in my effort to cap my hours. The failure taught me more than the diagnosis had. A governor regulates how much steam enters the engine. It has nothing to say about what the engine is driving. Capping my hours told me when to stop, But it said nothing about what to stop doing. I had been measuring work at the intake, the way one measures fuel, when the only measurement that matters is taken at the other end. The question was never how many hours I put in. The question is what remains after the hours are spent.
The question was never how many hours I put in. The question is what remains after the hours are spent.
Two kinds of work
Some of what I do ends the moment my engagement ends. A session, a workshop, a client meeting, a pitch, a class. It happens, the learner or the client judges it in the moment, and then it evaporates. You cannot pick it up next month and use it again. A good session is like a good meal. The people who were in the room remember it, and nobody can eat it twice.
A different kind of work leaves something behind. A case, a framework, a simulation, a template, an article. This work can be separated from the occasion that produced it, and it outlives that occasion. I can teach it again, adapt it, license it, publish it, hand it to a colleague, or build the next thing on top of it.
I noticed the difference in my own work by accident. A simulation I had been sketching in my head for years finally got built this spring because I had AI tools at my disposal. Within a few months, I had built six simulations. I had not increased my output because I spent more time. My productivity increased because my learning had started to compound. The first build was a trail of mistakes in design and development. I learned, I iterated, and then I did something I had never done with a piece of teaching material: I encoded what I had learned into a reusable skill, a set of instructions my AI tools could follow. The skill got better with each simulation, and the simulations got faster. I can now build one in a fraction of the time the first one took.
Contrast that with the enormous effort I pour into individual executive sessions and keynotes. AI has made those hours expand rather than contract. I can analyze a client’s business, rewrite the cases for their industry, speak their vocabulary, and produce decks more polished than anything I made three years ago. Each of those sessions produces one good audience experience. Then the effort evaporates, and the next client starts me from zero.
Same hours. Completely different afterlife.
The Carver and the Mold Maker
Look at your own week and you will find the same split.
A lawyer who drafts a superb memo for one client is a carver. He produces one beautiful object, and nothing survives beyond that client and that memo. A lawyer who turns the reasoning behind that memo into a clause library and a drafting method the whole firm draws on has made something very different. He has made a mold. The mold is not itself a memo, and that is precisely its value. A hundred memos can come out of it, drafted by him or by colleagues who never saw the original.
The advertising team that ships a brilliant campaign has produced a campaign. The team that extracts a repeatable creative method from it has produced a capability. The analyst who builds an elegant one-off model, the physician who reasons through a hard case alone, the salesperson who runs a masterful discovery call and writes nothing down: all of them are doing excellent work that evaporates.
Most professional weeks are dominated by carving, and we take this for granted. We use AI, we feel more productive while the work is happening, but we do not pause to consider what the work leaves behind. If we did, we would notice that it leaves no endowment. What we have at the end of the week is a satisfied client and a calendar entry.
Try an exercise. Open your calendar to last week and walk through it hour by hour, asking of each block what exists today that did not exist on Monday. Delivery does not count, since a carver’s week is made entirely of things delivered. The question is what still persists. I suspect you will find the answer is close to zero, and this has nothing to do with the quality of your work. It has to do with where the effort went.
Why this Matters More Now
You could have run the same exercise five years ago with the same result, and it would have mattered less. Building the reusable version was slow and expensive, the payoff was uncertain, and for most people the arithmetic did not justify it.
AI has rewritten that arithmetic. There is a paradox at the heart of these tools that took me a while to appreciate. AI is mediocre at inventing a strong structure out of nothing. Ask it for a case with no template and you get a plausible average of every case it has ever read. Give it a structure that already exists, however, and it becomes extraordinary at filling and extending it. Hand it a case template and a company, or a simulation engine and a new industry, and it will do in an afternoon what used to take months.
That asymmetry has two consequences. The first is that the return on time invested in durable structures has risen by orders of magnitude, while the return on polishing individual deliveries has not moved at all. If anything, it has reduced, because polish is now cheap. I spend more time perfecting decks with AI than I spent before I had AI, the decks are better, and my audiences are no more moved by them than they were, because a beautiful deck no longer signals anything. The second consequence is subtler. The mold is what makes AI’s output yours. Without one, the machine gives you the average of everything it has seen. With one, it gives you your version, in your form, carrying your judgment. The mold is where your taste, your judgment, and your experience become assets that can keep on giving.
So the new tools reward the person who owns a mold and penalize the person who hand-carves every piece. The question is no longer how much you work. It is whether what you make can be used again.
How to Find Good Problems
There is a second half to this argument, and it took me longer to see because at first it seemed to contradict the first half.
Last month I spent a day on the shop floor of a company that does CNC milling and injection molding for customers who need custom parts fast. I was there for the most evaporating kind of work I do, a keynote to the leadership team. A month later I have an academic article on integrated yield management for high-mix manufacturing, and an AI-native simulation called YieldForge that teaches yield management to MBA students and executives who will never set foot in that plant. There is an irony here that I only noticed while writing this article. That business earns its living by knowing that the mold is the asset and the part is the byproduct. I walked its floor for a day and came away, finally, understanding my own.
Something similar happened a few weeks ago, and this time I was on vacation, not even on a client engagement. I had dinner on a Friday at the impossible-to-get-in restaurant Bungalow in New York. After an excellent meal, I had the opportunity to chat with the chef, Vikas Khanna. He is gracious, and he shared a bit of his personal story and values with us. By Saturday the food was digested, but a question remained in my mind. How does a celebrity chef monetize a brand? Wolfgang Puck monetized distribution, through licensing and franchising. Gordon Ramsay monetized attention through reality TV. Khanna appears to be monetizing scarcity, keeping access to the restaurant tight while broadening access to the brand through merchandise like sauces. By Tuesday I had a finished teaching case, built with the case-writing skill I have honed over the past year, with a comparison of the three chefs and a set of dilemmas a class can argue about. That used to take me six months.
Put the two stories side by side and the lesson is the same. What made the durable artifacts possible was proximity to a person in the grip of a real dilemma. The plant supervisor deciding which of several urgent jobs gets the machine this hour. The chef weighing how far his name can be stretched before it thins. I could not have found these problems by reading about the industry. I had to be in the room.
This is the part that resolves the contradiction. The sessions and dinners and client work, the work that evaporates, are where the problems live. They are the mine. There is nothing wrong with spending time in the trenches. The waste comes from walking out of them with empty hands. Research, analysis, comparison, and drafting have become cheap. Access to interesting, unsolved problems has not, and if you teach, advise, sell, treat patients, or serve clients, you are already standing in a stream of them every week. The whole opportunity is in carrying what you saw out of the room and giving it a form that survives.
What it takes
Anyone can rent the models, so the models are not where the advantage lies. The advantage is in a set of skills and habits that AI cannot supply and that grow with use. Here is what I have found it takes.
It takes accumulated pattern recognition. Thirty-five years of this work means I am usually seeing the twentieth version of a problem rather than the first, and I can generally place it in its family within an hour. Experience cuts both ways, though. It also breeds selective perception. I sometimes see patterns that are not there, and I am tempted to force a new problem into a familiar shape because the familiar shape is ready to hand.
It takes getting out of the building, because deep insight does not come from reading about an industry. You must stand next to the person making the call at the moment they make it and notice the thing they have stopped noticing because they live with it every day. Then you must ask whether the problem generalizes. Scheduling under high mix and urgent demand turns out to be the same problem in specialty chemicals, clinical labs, and commercial print. A chef’s brand dilemma is the dilemma of anyone whose name has become a business.
It takes speed, because what you see will be seen by others, and because insight loses resolution with every day that passes. Three or four days out you still have your notes, and the texture is gone, and texture is what makes a case teachable rather than merely accurate. Most of the people who never produce anything had the insight and lost it on the drive home.
It takes subtraction, which runs against every instinct that made you good at the client work in the first place. A case containing everything you learned is unusable. Deciding what to leave out is the actual craft, and it is the part I still find hardest.
And it takes a repertoire of forms. Knowing what makes a case teachable, a simulation playable, a framework sturdy enough to bear weight. This is knowledge about containers rather than content, and it is why a person with a real insight and no form ends up with a LinkedIn post instead of an asset.
The Production Line and the Experience Curve
As I build more simulations and more cases, each takes a fraction of the time the one before it took, and the experience curve is steep. Every build leaves behind machinery that outlives the thing it was built for. A test harness that tells me whether skill beats luck in a game. A headless engine I can run thousands of times before a student ever sees it. Hard-won rules for what a decision screen has to show a player, and a feel for how many decisions a person can hold in mind before the game stops teaching anything.
This experience curve is the same manufacturing finding that unit cost falls by a steady percentage with every doubling of cumulative volume. The idea was never meant to apply to knowledge work, and it applies now only because I stopped making one-off things and started running a production line.
This also explains why so few people get to the good part. Judged alone, the first artifact is a poor investment. You pay for the equipment and receive one unit. Anyone who evaluates each piece on its own return quits after the second, which is exactly when the curve begins to bend. The arithmetic only works across the sequence, so the early builds must be funded based on conviction rather than evidence.
The institutions most of us work inside make this investment difficult. Billable hours pay for the memo, and they pay nothing for the clause library that would make the next memo take a quarter of the time. Teaching evaluations reward the session rather than the case that will be taught for a decade. Consulting contracts specify deliverables, but deliverables die with the engagement. Almost every incentive we have was designed for a world in which carving was the only option, and each of them militates against people who want to build a mold. The mold maker, for now, is usually self-funded.
One warning, because I have felt the pull. Once your machinery is good at a particular kind of thing, that kind becomes almost free and everything else starts to look expensive. Cheapness becomes a selection bias. You can find yourself building the same thing repeatedly because it is easy, rather than because it should exist.
Three questions before you begin
All of this reduces, in practice, to three questions I now ask before I spend serious effort on anything.
Will this exist next month? If not, cap the effort deliberately and do the work well within that cap. Some things genuinely should be a good afternoon and nothing more.
Could someone else use it if I handed it over? This is the harder test. Plenty of things that feel like assets are usable only by their author. Turning a private insight into something that can travel without you in the room takes foresight, and a form.
Am I with a person who is struggling with an unsolved problem? If so, pay attention, because problems are the scarce input now. Solutions have become abundant while good problems remain rare.
What I still have not sorted out
An artifact only pays if it travels. Build a simulation for your own classroom and never publish, license, or give it away, and you have built yourself a very expensive lesson plan. Distribution is real work, and it is not the part I enjoy.
Some effort still evaporates, and I have decided to let it. Having seen what a session can be, I cannot give a room of executives the older version. My conscience will not let me teach something generic when I know what the customized version does for the people in the room, even though I am not getting paid for my customization effort. I spend those hours because the work is worth doing well, and I have stopped pretending otherwise.
The scarcity does not disappear. It moves. When a case takes four days instead of six months, cases stop being scarce, and the hard part becomes deciding which cases deserve to exist at all. Nothing here exempts artifacts from the inflation that created the original problem. The mold makes the tenth piece cheap, and it says nothing about whether the world needs a tenth piece.
And there is a tension I have not resolved. Knowing that a dinner conversation can become a case by Tuesday makes the world look full of things worth making, and the well of worthwhile work I described last month gets deeper rather than shallower. Choosing what to work on turns out to make it harder to limit how much you work. You need both disciplines, and the second one fights the first.
I return to my wife’s question. I think I have a better answer for her. I was busier because I was feeding an engine that drove nothing, and cheaper fuel only made me feed it faster. The governor was the wrong instrument. The right one is a ledger, kept at the end of every week, of what now exists that did not exist before. I would rather spend my remaining working seasons on things that outlast the afternoon they were made for. And I can now tell the difference before I begin, which is more than I could say a month ago.
Prof, One wrinkle: the mold itself may not stay scarce for long. Once a framework, template or simulation is legible, AI makes imitation cheap too. The compounding advantage may come from staying close to new problems and continuously updating the mold with fresh edge cases. Durability then depends on the learning loop, not simply on turning work into an artifact.