The Ambition Gap: When AI Makes Building Cheap

A data report from Linear made the rounds this month, and one number in it should stop every builder cold. Teams that connected a coding agent tripled their weekly pull requests over two years, from 21 to 65. Teams that did not went from 8 to 10. AI now authors close to half of all the issues those teams file, up from about 1 in 1,000 two years ago. Making software, the thing that used to be the hard part, is collapsing in cost in front of us.

So here is the question almost nobody is asking while they celebrate the throughput. If building just got five times cheaper, why are most people producing roughly the same things they produced last year, only faster?

That gap is the whole story. When the cost of making something falls off a cliff, your results stop being limited by what you can build. They start being limited by what you are willing to attempt. Skill was your ceiling for your entire career. It quietly stopped being your ceiling, and most people have not updated their goals to match. I call the space between what you could now attempt and what you actually attempt the ambition gap, and right now it is the single biggest thing holding good builders back.

I have watched this in my own work and in dozens of founders I talk to. The tools got absurdly powerful and the plans stayed the same size. This is the post I wish someone had put in front of me the first time a model did a week of my work in an afternoon and I used the saved week to do more small work.

The productivity paradox is an ambition problem

Start with the strangest fact in business right now. About 91% of firms report using AI. Around 89% of managers report no measurable change in productivity. Only about 6% of executives say they can point with confidence to organization-wide return on their AI spend. Adoption is nearly universal and the payoff is nearly invisible. Economists have started calling it the new Solow paradox, after the old line that the computer age was everywhere except in the productivity statistics.

Now hold that next to the task-level numbers, because they tell the opposite story. Depending on the study and the task, individual workers get somewhere between 14% and 55% faster with AI help. Developers routinely say they feel twice as fast. So the person feels faster, the task is faster, and the company sees nothing. Where does the speed go?

Two places, and both matter. Some of it backs up at the human gates, the reviews and approvals and judgment calls that AI did not remove, so a flood of cheap output just piles up behind a reviewer who is now the bottleneck. That part is real and I have written about the downstream cost of cheap output before. But the bigger leak is quieter and it is the subject of this post. People take the time AI gave them back and spend it doing more of the same small work. The goal never changed. Only the pace did.

Think about what that means. If you hand someone a machine that does their work in half the time, and their output looks identical except there is twice as much of it, they did not get more ambitious. They got faster at their existing ambition. And a company full of people getting faster at their existing ambitions produces exactly what the paradox shows: a lot of motion, more artifacts, and flat results.

Reports keep confirming the pattern from the other side too. A large majority of enterprise AI pilots, by some counts 95%, never reach production at all. And where companies do capture the speed, surveys find they mostly use it to demand more of the same output from the same people, turning an eight-hour workload into a bigger eight-hour workload rather than pointing the freed capacity at something that was previously impossible. The surplus is real. Almost nobody is spending it on a bigger target.

That is the tell. The productivity paradox is usually explained as a measurement problem or an adoption problem. It is neither. It is an ambition problem wearing a productivity costume. The capability jumped and the aims did not. You can see it in your own week if you look closely. The tool got faster, the to-do list got shorter, and the plan sitting at the top of it never changed shape. That is the paradox in miniature, one person at a time, and it is the reason the fix is not another tool or another hour saved.

The moved constraint

To see why flat results follow cheap execution, borrow an idea from a factory. In 1984 Eliyahu Goldratt wrote a business novel called The Goal that introduced the theory of constraints to a wide audience. Its core claim is deceptively simple. Every system, a plant, a team, a career, is limited in its output by exactly one binding constraint at a time. There is always a slowest step, and the whole line moves at the speed of that step.

The part people forget is the second half of the claim, and it is the part that matters here. Improving anything that is not the binding constraint produces no gain. If a machine that is not the bottleneck runs faster, all it does is pile up inventory in front of the real bottleneck. Goldratt put it bluntly: time spent optimizing a non-constraint is not a small win, it is waste. You get the appearance of improvement and none of the result.

Now map that onto a builder. For your whole working life, your binding constraint was execution. You could imagine more than you could make. The bottleneck was skill and hours: your ability to write the code, design the thing, produce the pages, ship the feature. Ambition sat comfortably behind that wall, because it did not matter how big you dreamed when you could only build so fast. The slow step was your hands.

AI moved the constraint. Execution is no longer the slow step. The machine widened that valve so far that it stopped being what limits you. And by the logic of constraints, the moment a bottleneck opens up, the limit relocates to the next narrowest point. For most builders, that next point is ambition, the size and boldness of what they choose to aim the new capability at.

This reframes almost every piece of advice floating around about the AI era. The prompt-engineering tips, the ten-tools-you-must-try threads, the race to shave another few minutes off a workflow, all of that is optimizing execution. It is real work on a valve that is already open. It feels productive because motion always feels productive. But if execution is no longer your binding constraint, that work produces no change in your results, in exactly the way Goldratt warned. You are speeding up a machine that is not the bottleneck, and the output just backs up behind the constraint you refuse to touch.

The ambition gap is what that refusal costs you. It is the distance between the ceiling the tools now allow and the ceiling you actually operate under. And because the tools improved silently and continuously, that distance has been widening under your feet without any single moment that forced you to notice.

The two ceilings that just swapped

Picture two horizontal lines above your head. The lower one is your skill ceiling, the most you can currently build with your hands and hours. The higher one is your ambition ceiling, the biggest thing you have actually let yourself aim at. Whichever line is lower is the one you hit. That is the binding constraint made visual.

For most of history, for almost everyone, the skill ceiling was the lower line. You wanted to build more than you could. The dreamer with no ability to ship is the oldest character in the room. So skill was binding, and the entire self-improvement industry was correctly built around raising it: learn faster, work harder, get better at the craft. Push the lower line up and your results rose with it.

AI shoved the skill ceiling up through the roof, past the ambition ceiling, in the span of about two years. And here is the trap. When the two lines cross, the binding constraint switches from one to the other, but nothing about the switch announces itself. You still feel the same. You still have the same goals written down. You just quietly stopped being limited by the thing you spent your whole life training to fix, and started being limited by the thing you never trained at all.

The reason this is worth naming is that most of us are still running the old playbook against the new constraint. We keep pushing on skill, because pushing on skill is a habit thirty years deep and it has a clear method. Pushing on ambition has no course, no certification, no morning routine. It is uncomfortable and unmeasured. So we default to the familiar move, and the familiar move is now the wrong one. This is the same failure mode I described in the difference between motion and progress: the work feels like progress precisely because it is the work you know how to do.

Why you bank the savings instead of betting them

When AI hands you back a week of time, you have two choices with it. You can bank the savings, meaning you keep the same goal and pocket the surplus as speed and slack. Or you can bet the savings, meaning you take the freed capacity and point it at a bigger goal that was out of reach before. Banking is the default. Almost everyone banks. It is worth understanding why, because the reasons are not laziness, they are structural.

The first reason is anchoring, and it is the strongest. Your sense of what is a reasonable goal was set by your old skill ceiling. You spent years calibrating what is ambitious but achievable, and that calibration is now stale, tuned to a version of your abilities that no longer exists. When you sit down to plan, the number that feels right is the number that felt right last year, because that is what your gut was trained on. The gut has not been retrained. So you aim where you always aimed, and it feels responsible rather than timid.

The second reason is that output is measurable and ambition is not. You can count pull requests, posts, features, calls. You cannot easily count whether you are attempting something big enough. So when you look for evidence that you are doing well, you reach for the countable thing, and the countable thing rewards banking. More output looks like more progress on every dashboard you own. A bigger, riskier bet often looks like less activity for a long time before it looks like anything at all. The metrics quietly push you toward the small.

The third reason is loss aversion pointed at your ego. A bigger goal carries a real chance of visible failure. Staying at your proven level carries almost none. When execution was expensive, this caution was rational, because a failed big swing cost you months of scarce build time. Now a failed swing costs you a fraction of what it used to, since the building is cheap. The price of attempting fell, but the fear did not, so we keep paying an old premium for safety we no longer need.

Here is the research that should embarrass all of us into raising the aim. Edwin Locke and Gary Latham spent decades studying goal setting, and their central finding is one of the most replicated results in the field. Specific and difficult goals produce dramatically higher performance than easy or vague ones. In Locke’s work, people with hard goals outperformed people with the easiest goals by well over 250%. The catch is that when left to choose their own targets, people reliably set them too low. We underaim by default, and we have been doing it since long before AI made underaiming this expensive.

The ambition anchor

Put a name on the specific failure so you can catch it. The ambition anchor is the goal you are carrying that was sized for the person you were before the tools. It is the ten-times tool pointed at the one-times goal. You upgraded your capability by an order of magnitude and left the target sitting where it was, and now you are using a machine that can move mountains to move the same molehill you were always going to move, just faster.

You can hear the anchor in the language people use. They say things like, I want to ship this feature by end of quarter. Reasonable, except that with an agent you could have shipped it in a week, so the quarter-long plan is anchored to a pace that no longer applies, and the extra eleven weeks are being spent on nothing bigger. They say, I want to grow the newsletter to five thousand. Reasonable, except that the reason the number is five thousand is that five thousand felt like a stretch when writing each issue took a full day, and it no longer does. The goal did not grow when the constraint did.

The anchor is sticky because it is invisible. Nobody writes a goal and labels it capped by my 2024 abilities. It just feels like the right size, and feelings of rightness are exactly what anchoring produces. The only way to break it is to run a deliberate check, which I will give you in the Monday section: take your current top goal and ask whether it would have been achievable with your old skills plus more hours. If the answer is yes, you did not set a goal for the new world. You set a goal for the old one and let the machine sprint at it.

This is also why knowing what to learn in the AI era matters more than it looks. The point of new skill now is not to out-execute the machine at the thing it does well. It is to expand the set of problems you can even conceive of aiming at, so that your ambition ceiling has somewhere to rise to. Learning becomes a way to see bigger targets, not a way to build faster.

What raised ambition actually looks like

The abstract argument only lands if you can see what betting the surplus produces, so look at the teams that did it. The pattern in the numbers is not that they hired armies. It is the opposite. They aimed a tiny group at a target that used to require an army, and let the cheap execution close the gap.

Midjourney generated around 500 million dollars in revenue with roughly 107 people, about 4.7 million dollars per employee, with no outside venture funding, no sales team, and no paid marketing. Cursor crossed two billion dollars in annualized revenue with roughly 300 people, near 6.7 million per person. Gamma reached 100 million in annual recurring revenue with about 50 people and stayed profitable for years. The median private software company runs near 130 thousand dollars of revenue per employee. The AI-native leaders are running at three million and up. That is more than a twenty-fold difference in revenue density, and it did not come from those teams typing faster.

It came from where they pointed the surplus. A traditional company would have taken cheaper execution and used it to lower costs on the same product. These teams used it to attempt a product and a market that a group their size had no business attempting, and the tools made the attempt survivable. Investors have started pricing this directly. Sequoia now underwrites what it describes as the ability of tiny teams to produce outsized output, and it is repricing companies around the size of the swing a small group can take rather than the size of the group.

Even the loudest voice in the hardware of all this frames it as an ambition question, not a productivity one. Jensen Huang has been telling leaders that as long as you have greater ambition than your organization can currently handle, AI only creates work rather than removing it. His line is that the backlog of ideas and aspiration is enormous, and that automating the task of building lets you attempt more, not do the same amount with fewer people. Strip away the keynote framing and it is the same claim as this post from the top of the market. The constraint moved to ambition, and the winners are the ones treating it that way.

None of this requires being Midjourney. The mechanism is fractal. A solo founder who used to build one product can now attempt three, or attempt one that is ten times harder. A marketer who used to run one campaign can attempt a category-defining one. The move is always the same: notice that execution stopped being the wall, and walk through the space that opened up instead of pacing faster in the old room. It is the same logic behind the one-person company that keeps only an incompressible core of human judgment and points cheap execution at everything else.

Five surpluses, banked or bet

Ambition sounds vague until you make it operational, so break the surplus AI hands you into concrete pieces and look at the banking move and the betting move for each. Every row is a real fork you face, probably this week. The banking column is what you will do by default. The betting column is what raising the ambition ceiling actually requires.

Read the banking column top to bottom and notice how sane every entry sounds. Ship sooner, cut cost, stay safe, polish. None of it is a mistake in isolation. It is the collective effect that is the trap: a person who banks every row is fully occupied, visibly productive, and going exactly as far as they went last year. The betting column feels riskier line by line, and it is the only column that moves the ceiling.

There is a version of this that goes wrong, and I want to name it before the contrarian section so it does not read as a blanket cheer for bigger. Betting the surplus on the wrong target burns it faster than banking ever would. Cheap execution aimed at a bad idea just produces a bad idea at scale. Which is the honest hinge of the whole argument, and where the next section starts.

The contrarian take: the scarce thing is nerve

Everyone has a favorite answer for what becomes scarce and valuable when AI can build anything. The usual candidates are skill, taste, and judgment. I have argued for taste and judgment myself, and they matter. But watch what actually stops people, and a different answer shows up. The thing in shortest supply is not the ability to tell a good target from a bad one. It is the nerve to commit to a big one and own the outcome.

Look at the behavior. People are not failing to raise their ambition because they cannot identify a bigger goal. Ask any competent builder what they would attempt if building were free, and they can name it in about four seconds. They already know the bigger target. What they lack is the willingness to point at it out loud, put their name on it, and accept that a public swing can publicly miss. That is not a taste problem or a skill problem. It is a nerve problem, and nerve is the one input none of the tools supply.

This is why so much AI energy pours into optimizing execution. Optimizing execution is safe. It is measurable, it is legibly hard work, and it never requires you to declare a goal that might make you look foolish. It is the productive-feeling way to avoid the actual constraint. A person can spend a year mastering every tool, shipping constant output, and looking industrious the entire time, all while never once raising the ceiling that determines their results. That is the comfortable trap, and it is exactly the efficiency trap seen from the inside of a single career: getting very good at doing things that were not worth scaling.

Now the honest counterweight, because nerve without discernment is just expensive flailing. Raising ambition is only the right move if you can pick a target worth the swing, and that is where taste and judgment come back in. The full move is two steps, not one: aim higher, and aim well. Nerve gets you to attempt something big. Taste keeps you from aiming the cheap execution at a large mistake. The reason I am putting weight on nerve here is not that judgment stopped mattering. It is that judgment is the part everyone is already working on, and nerve is the part almost nobody will admit is the real blocker. The scarce input is the one you are least willing to name.

The ambition map

Two forces set where you land, and they are independent. One is how much execution power you have, which for most builders is now high and rising whether they use it well or not. The other is how big a target you are actually aiming at. Cross them and you get four places to be, and only one of them is where the results are.

The old failure was the top left, the frustrated dreamer with grand plans and no way to build them. That quadrant is emptying out, because the tools now hand almost anyone the execution to leave it. The new failure is the bottom right, and it is crowded. Huge execution power aimed at a small goal. Enormous capability, spent on the same modest target, producing the faster hamster wheel that the productivity paradox measures across the whole economy. It is the most comfortable place on the map, because you are busy, capable, and shipping, and it is a dead end dressed as momentum.

The entire move this post is arguing for is a single vertical step from the bottom right to the top right. Not more execution power, you already have that. A bigger aim to point it at. The horizontal axis is handled for you now. The only axis you still control is the one going up.

What to do Monday morning

None of this matters if it stays a mindset, so here is the audit I run on my own goals, and it takes about twenty minutes. Do it on your single most important current objective, the real one, not the tidy one you would say in a meeting.

First, run the old-ceiling test. Write down your top goal for this quarter, then ask one question: could I have hit this with my skills from two years ago plus more hours and more people? If the answer is yes, the goal is anchored to your old skill ceiling. It is a banking goal. It survived the arrival of the tools unchanged, which means it never accounted for them. Circle it. It needs to grow.

Second, run the ten-times question, and answer it honestly rather than reasonably. If execution were completely free, no time cost to build anything, what would you actually attempt? Write that down too. The gap between this answer and the circled goal above is your ambition gap made concrete, in your own handwriting. The point of the exercise is that execution is now close enough to free that the honest answer and the reasonable answer should be converging, and for most people they are still miles apart.

Third, reallocate rather than accelerate. Take the single biggest chunk of time AI has actually freed up in your week, and refuse to spend it on more of your current work. Point it at the harder target from the ten-times question, even a small slice of it. This is the practical side of the shift from producing to directing: the instinct will be to use the freed time to get further ahead on what you were already doing. That instinct is the banking reflex, and beating it once, deliberately, is the whole practice.

Fourth, change what you measure. If your only metrics are output counts, you have built a dashboard that rewards banking and punishes betting, and you will drift toward the small without ever deciding to. Add one outcome metric that only moves if you attempt something bigger, revenue from a new line, users in a market you did not serve, a result you would not have dared to write down last year. Watch that number instead of the activity ones.

Fifth, commit to one bigger bet out loud. Locke and Latham are clear that goals work best when they are specific, hard, and committed to. Pick one target from the top-right of the map, make it concrete, put a date on it, and tell someone who will remember. The public commitment is not motivation theater. It is the mechanism that overrides the loss aversion keeping you small, because now the cost of quietly shrinking the goal is that someone will notice.

And keep one running test on the calendar. Every quarter, look at your new plan next to the last one. If this quarter’s goal is basically last quarter’s goal plus twenty percent, you banked. A world where your execution power is compounding should not produce goals that grow in tidy twenty percent steps. It should occasionally produce a goal that makes you slightly nervous to say. If nothing on your list makes you nervous, your ambition is still anchored, and the tools are still doing small work for you at a very impressive speed.

FAQ

Is the ambition gap just a fancy way of saying dream bigger? No, and the difference is the whole point. Dream bigger is a motivational slogan with no mechanism. The ambition gap is a claim from the theory of constraints: your results are capped by exactly one binding constraint, that constraint used to be execution, and AI moved it to ambition. This tells you precisely why working harder on your skills now produces no gain, which no amount of dreaming bigger explains. It is a diagnosis, not a pep talk.

How is ambition different from taste or judgment? Taste is knowing what is good. Judgment is knowing which option to pick. Ambition is how big a target you point at and whether you will own the swing. They are different constraints, and right now they bind in sequence: you need the nerve to aim higher, and then the taste to aim well. The reason I put weight on ambition is that taste and judgment are the parts everyone is already trying to improve, while ambition is the part people quietly leave anchored.

Does raising ambition just mean doing more work? The opposite. More work is the banking move, doing more of the same at a higher pace. Raising ambition often means doing less of the current work so you can point the freed capacity at one harder thing. Output usually goes down in the short term when you bet correctly, because you stopped spending the surplus on visible busywork. That temporary dip is a feature, not a warning sign.

Won’t everyone raise their ambition, so the bar just resets? The whole argument is that they will not, and the productivity paradox is the evidence. With near-universal AI adoption, most people and companies are showing no change in results, which only happens if the surplus is being banked rather than bet. The bar is not resetting. The vast majority are staying at the bottom right of the map. That is exactly why the small number who step up have room to run.

How do I know if my current goal is anchored to my old ceiling? Run the old-ceiling test. Ask whether you could have achieved the goal with your skills from two years ago plus more hours and more hands. If yes, it is anchored, because it does not require the new capability at all. A goal built for the new world should be one that was genuinely out of reach before the tools arrived and is reachable now.

What if execution really is not cheap yet in my field? Then execution is still your binding constraint, and you should keep raising your skill ceiling. The framework holds either way. The move is to always work the actual constraint, and to keep checking which one it is. The risk is not that you raise ambition too early. It is that execution quietly gets cheap in your field, as it has in most, and you keep polishing skill out of habit long after it stopped being the wall.

Is more output ever the right call? Sometimes, when volume is the actual goal, more output is genuine progress rather than banking. The test is whether the extra output moves an outcome you care about or just fills a dashboard. If shipping twice as many features raises revenue or reach, that is a real bet. If it just produces more features that sit there, you banked and called it productivity. Watch the outcome, not the count.

Where do tiny teams like Midjourney actually fit in this? They are the clearest evidence that betting the surplus works. A group of around a hundred people producing hundreds of millions in revenue, or fifty people running a profitable company at a nine-figure run rate, did not get there by executing faster on a normal-sized goal. They pointed a small team at a target that used to require a large company, and cheap execution made the attempt survivable. That is the top-right of the map at the scale where it is easy to see.

The uncomfortable summary is short. Making got cheap, so your ceiling stopped being what you can build and became what you dare to attempt. Most people are spending the surplus going faster at the same small thing, which is why their output rose and their results did not. The move is not another tool or another skill. It is to raise the aim, out loud, at something that was out of reach a year ago and is not anymore. For more of how these pieces fit together, the founder operating system and its AI-age update are where I keep the rest of the map, and the reps problem covers the skills worth protecting even as execution gets cheap.