AI Made Execution Cheap. Decision-Making Is Becoming the Moat.

For most of history, building things was expensive. Writing software was expensive. Producing an advertisement was expensive. Designing a new product, running a piece of research, even putting together a decent presentation, all of it consumed real time, real money, and real people. So companies did the rational thing: they got very good at execution. They hired better engineers, better designers, bigger marketing departments.

They moved faster and shipped more, on the fairly reasonable assumption that if you could execute better than everyone else, you had an advantage.

For a long time, that was true. I'm not sure it will remain true for much longer.

AI is doing something bigger than making people more productive. It is quietly moving the bottleneck. Execution is becoming abundant. Decision-making is not, and that gap is where I think the next decade of competitive advantage actually gets built.

Before they flew, they simulated

In 1901, the Wright brothers had a problem. Their gliders weren't behaving the way the aerodynamic data available to them said they should. They could have kept building increasingly expensive, increasingly dangerous aircraft and learning through failure, the way most of their competitors did.

Instead, they built a wind tunnel, a small one, nothing dramatic, and started systematically testing wing shapes under controlled conditions, generating their own aerodynamic data before committing to another aircraft.

Before trying to fly again, they built a cheaper environment in which they could be wrong.

That distinction is easy to skim past, but it's the whole point. The Wright brothers' breakthrough wasn't that they became better at building aircraft. It's that they became better at deciding what aircraft should be built. We've since generalized that idea almost everywhere else in engineering. Bridges get simulated before they're constructed. Aircraft manufacturers model aerodynamics long before a prototype exists. Car companies crash-test vehicles instead of putting people inside them on faith.

Chip designers simulate circuits. Factories increasingly run as digital twins before a single machine is installed.

There's a fairly obvious reason for all of this: reality is an incredibly expensive place to discover that you were wrong.

And yet there's one enormous category of decisions where businesses still routinely do exactly that: decisions involving people.

The bottleneck moved

Picture a product launch a few years ago. The website took weeks. The campaign needed an agency. The copy passed through several sets of hands before anyone was comfortable running it. Producing five visual directions cost actual money, and building the product itself could take months. Today you can ask an AI system for twenty headlines before lunch. Twenty landing pages. Twenty ad concepts. Twenty onboarding flows. Twenty pricing variants. Twenty product ideas.

And your engineering team can prototype several of them at a speed that would have sounded absurd five years ago.

That's obviously a win. But it's not the full picture, because the hard part hasn't gone anywhere. You still have to choose. Which headline? Which product? Which market? Which feature? Which price? Which campaign? Which customer? Which investor? Which partner? AI can hand you another hundred options without telling you which one deserves to exist. Here's the ironic part. In some ways, all that speed actually makes the underlying problem worse.

When generating ideas is practically free, the real scarcity shifts to choosing between them. The bottleneck stops being execution. It becomes judgment.

Almost every business decision eventually becomes a decision about people

I think we sometimes make business sound more complicated than it actually is. We talk about product strategy, marketing strategy, capital strategy, sales strategy, pricing strategy, hiring strategy, brand strategy, different departments, different spreadsheets, different vocabulary for what is, underneath, the same question asked in different rooms. Follow almost any consequential business decision far enough and you eventually run into a person.

A product decision becomes: will somebody actually use this? A pricing decision becomes: will somebody believe the value is worth the cost? A marketing decision becomes: will somebody notice this, understand it, and care? A brand decision becomes: what will somebody believe about us after seeing this? A sales decision becomes: will a buyer trust us enough to take the risk? A hiring decision becomes: will this person join, perform, and work well with everyone else? Even a fundraising decision reaches the same floor eventually: will an investor believe the potential outcome justifies the risk of putting capital into this company?

Different people, different incentives, different rooms, same underlying problem: you're trying to predict how humans will react to a decision that hasn't happened yet. Customers are just the most visible version of this. Investors are people. Partners are people. Procurement teams, employees, executives, consumers, all people. A company survives because enough of them, repeatedly, decide to say yes instead of no.

Everything works. And people still leave.

Imagine spending six months building something. The technology works, the interface looks good, the offer seems reasonable, and the engineering team has genuinely done its job. Then people show up. They hesitate. They misunderstand something you thought was obvious. It’s not that they hate it. They’re just not sure about you yet. Maybe the price stings a little. Maybe they like what you’re selling but can’t quite talk themselves into buying it right now.

Or maybe the person actually using your product thinks it’s brilliant, but the one signing the checks isn’t convinced. And sure, your ad might grab attention, but it’s probably saying the wrong thing. Everything technically works, and people still leave.

This is where most of our current tools become strangely limited. Analytics can tell you where someone left. A heatmap can show you where they clicked. Conversion data can tell you Variant B beat Variant A by eleven percent. All that data is helpful, don't get me wrong. But here's the catch: by the time you get it, the decision's already been made. The traffic's been bought. The campaign's already run. The product's already built. You've already had that investor meeting.

Reality has already sent you the bill. And more importantly, knowing what happened still doesn't tell you why it happened.

Did they just not get it? Did something along the way make them second-guess you? Was the whole process just too much mental work for what you were asking them to do? Or did another option just feel like the safer bet?

And here's the kicker. Was it really the price? Or did you just never make the value clear enough for price to even enter the conversation? Because those are two completely different problems. But they can both look exactly the same in your analytics: USER_LEFT_PAGE.

A wind tunnel for decisions

So what does better decision-making actually look like? I don't think the answer is finding executives with superhuman intuition, and I definitely don't think it's asking an LLM "what would customers think about this" and treating the reply as market research. The more interesting possibility is changing the economics of testing decisions themselves, going back to the Wright brothers and, in effect, building the wind tunnel.

What if you didn't have to wait until everything was already done to figure this out? What if you could test the waters before you actually rolled the dice?

Picture this. Before you sink half a million into a campaign, you could know how different audiences will react. Before you redo your entire onboarding, you could pinpoint exactly where people get stuck and why. Before you change your pricing, you could understand how trust, value, and risk shift for different kinds of customers. And before stepping into a new market, understand how the exact same offer lands with people who see the world very differently than your usual crowd.

None of this is because a simulation can magically tell you the future. It can't. The Wright brothers' wind tunnel wasn't the sky either. It was useful because it gave them a cheaper environment for testing assumptions before they had to confront the sky for real. That's the distinction worth holding onto.

This is, ultimately, the problem we're trying to solve with Aetherya. Not "AI personas." We're not building another chatbot that fakes being a human decision-maker with a made-up profile. The larger idea is cognitive simulation: building computational models of people and populations that can be exposed to a decision before that decision reaches the real world.

Models that don't just generate an answer but carry state, that can hesitate, whose trust can rise or fall, where new information creates cognitive load, where an objection can surface and new evidence can change a position, and where different people can look at exactly the same thing and interpret it differently.

So with Thymos in Aetherya, we're building a cognition-state model around that. We're modeling the underlying reasons for a reaction, not just treating every simulated person as another faceless LLM. We're not trying to predict customer behavior. We want to see how a decision might affect different people, understand why they'd react that way, spot the weak assumptions, and figure out what to test next.

Everyone gets the same leverage

There's another side to this that's easy to miss. You're not the only one with access to AI. Your competitors have it too. They can generate copy, generate code, research markets, produce prototypes, automate operations, the same list you can. Simply "using AI" is probably not much of a moat by itself. Eventually it becomes the equivalent of using the internet: expected, not differentiating.

The more interesting question is what happens once two companies have roughly comparable execution capability. Company A can produce ten campaign directions in a day. So can Company B. What separates them at that point is the quality of the decision about which one gets the budget. Company A can generate fifty product features; so can Company B. The advantage belongs to whoever can identify which feature actually moves customer behavior in the intended direction.

The faster execution gets, the more expensive bad judgment becomes, because AI lets you execute a bad decision just as quickly as a good one.

Speed without direction is not an advantage

Startups repeat a phrase constantly: move fast. I agree with it, but speed only helps when the feedback loop underneath it is good. Otherwise you're just arriving at the wrong destination more efficiently than you used to.

AI has dramatically increased the velocity side of that equation. What I think comes next is an arms race around feedback: who can understand the consequences of a decision fastest, test the largest number of plausible alternatives, surface an objection before a customer does, catch a bad assumption before six months get spent building around it, and change direction while changing direction is still cheap. That company isn't just executing faster.

It's learning faster, and learning speed compounds in a way raw output never does.

The companies that win may simply be able to be wrong more cheaply

I keep coming back to this idea: the purpose of simulation isn't to eliminate uncertainty. That would be impossible. It's to move some of the failure away from reality and onto something cheaper. A failed simulation costs almost nothing. A failed product launch does not. A simulated customer rejecting your pricing is a data point; ten thousand real customers rejecting your pricing is a business problem.

A simulated audience misreading an advertisement is something you fix tonight; a real audience misreading it after you've spent €1 million distributing it is a post-mortem.

There will always be a point where reality has to take over. Human research will still matter. Actual customers will still surprise us. Markets will remain messy, and people are probably the hardest systems we could ever try to model. But we didn't stop using weather models because the weather stayed uncertain, and we didn't stop using engineering simulations because materials occasionally behave unexpectedly. We use models because making decisions with imperfect evidence beats making them with none.

There's no good reason to treat human behavior differently.

Decision-making is becoming the moat

For the last decade, a lot of technology was built to help companies execute. The next layer is going to be about helping them decide, not replacing the person making the call, but giving them a better environment in which to make it. AI made producing possibilities cheap. What's missing is a smarter way to figure out which of those possibilities is actually worth bringing to life.

Ignore the departments, the dashboards, the funnels, the strategy decks. Get rid of all that noise, and most businesses are doing something surprisingly simple. They make decisions. Those decisions affect people. People react. And those reactions determine what happens next.

The company that understands that reaction before everyone else wins.

The Wright brothers built a wind tunnel because crashing was an unnecessarily expensive way to learn aerodynamics. I suspect launching decisions straight into people and waiting to see what happens will eventually look just as primitive. That's the future I want Aetherya to help build, a place where organizations can test a decision before real people experience its consequences. Not because we can know tomorrow with certainty, but because we can walk into it considerably less blind.

---

Reality is an incredibly expensive place to discover that you were wrong