Your Next Competitor Has No Capital

Here is the thing nobody is pouring trillions of dollars into: an app that shortens your emails. The chatbot everyone is staring at — the thing that drafts your posts and generates pictures on demand — is the smallest, most visible sliver of what is actually happening. Deciding whether AI is a bubble by looking at it is like deciding whether the internet was overhyped in 1996 by judging how good it was at sending birthday messages.

The instinct to call it a bubble is understandable. We are trained — investors most of all — to read a chart that goes straight up as a warning. Every historical model we own says "this reverts." But those models were built for a world that moved in straight lines, and almost nothing about this moment is a straight line. When the underlying thing isn't a faster horse but a different category of animal, the old reflexes misfire. The chart looks like 1999. The substance does not.

You're pricing the wrong object

Here is the category error. If you value AI as a faster search engine or a better writing assistant, the multiples look insane and the bubble talk makes sense. But that is not what the capital is chasing. The deeper shift is that AI is driving the cost of knowledge toward zero. Almost everything in our economy is built on the opposite assumption — that expertise is scarce and expensive. Education, professional services, the entire org chart, the premium you pay for the person who knows the thing you don't: all of it priced around that scarcity. Take the scarcity away and you are not looking at a better tool. You are looking at a different economy. You don't capture that with a P/E ratio.

The revolution is operational, not conversational

The reason the bubble debate keeps missing is that it confuses the consumer surface of AI with the engine underneath. The chatbot is a surface. It is where most people touch AI, so it is what they assume AI is. But the value was never in the conversation. It is in intelligence wired into how organizations actually run — the decisions, the workflows, the unglamorous back-office machinery that moves money, goods, and risk around every day.

Picture the difference. A chatbot is a smarter way to ask questions about the work. An AI agent does the work — quotes priced, claims triaged, shipments rerouted, exceptions resolved — with a human setting the goals and the guardrails rather than performing every step by hand. The chatbot talks; the agent acts. One is a feature. The other is a new cost structure. Investors can smell that difference even when they can't yet see it on a screen, and it is most of what the capital is really betting on.

That second thing is the part I spend my days inside. We build the orchestration layer that puts AI agents to work inside operationally heavy, regulated businesses — aviation, finance, logistics — where "move fast and break things" is not on the table and a wrong answer has real consequences. From that seat, the chatbot looks like the demo. The real product is an agent that does the work, inside a process, accountably, at a marginal cost of roughly nothing. Most of it is not visible yet, which is precisely why it is underpriced in the conversation and overpriced in the panic.

The bottleneck was never the model

If you take one thing from this, take this: in enterprise AI, the model is almost never the hard part.

Here is what I see on nearly every deployment, and it is consistent with what most studies of enterprise AI adoption find: a large share of the organization quietly works against the rollout. Not out of malice — out of self-preservation. The technology works in the demo. Then it meets the organization: the team that doesn't trust it, the process nobody ever documented, the compliance officer who needs to know who is accountable when the system acts, the integration with the platform from 2009 that everyone forgot the password to. That gap — between a capability that works and a capability that is adopted and trusted to act — is where the overwhelming majority of corporate AI quietly dies.

The bubble debate has no vocabulary for this, because it is not a technology question at all. Models are converging and commoditizing fast; the distance between the best and the second-best narrows every quarter. The durable advantage is no longer "who has the best model." It is "who can get intelligence into the bloodstream of the organization, trusted enough to actually do the work." That is an orchestration, governance, and change problem. It is deeply unglamorous. It is also where the value lives.

And here is the part incumbents should find encouraging, if they're honest enough to hear it: the fact that this is hard is the whole opportunity. If adopting AI were as simple as buying a license, it would confer no advantage — everyone would do it overnight and the edge would evaporate. The difficulty is the moat. The companies that build the muscle to put intelligence into their operations safely, and to bring their people along with them, are building something a competitor cannot copy with a purchase order. Most of your rivals will treat this as an IT procurement. The ones who treat it as an operating transformation will pull away.

What does that work actually look like? Less than you'd hope from a vendor demo, more than a license. It's mapping where decisions really get made versus where the org chart says they do. It's deciding what a machine is allowed to do unsupervised and what stays in human hands. It's wiring the AI into the systems you already run instead of around them, and giving your people a reason to trust it rather than route around it. None of that fits inside a pilot. All of it is where the return is.

Kodak didn't lose on technology

There is an example I keep returning to. Everyone assumes Kodak died because it missed digital. Read the actual record and you find close to the opposite: Kodak saw it coming, was nearly paranoid about it, and out-invested its rivals in R\&D and engineering. It had better technology than almost anyone in the room. It lost anyway — because the rules of the game changed, and Kodak kept playing the old game brilliantly.

That is the exact risk facing most companies right now, and it has nothing to do with being slow on the technology. The risk is being fast on the wrong thing. Building a marginally better chatbot. Buying a marginally nicer dashboard. Running a dozen impressive pilots that never once touch the P\&L. Meanwhile the real game — expertise at near-zero marginal cost, woven into operations — shifts underneath you. You can win every battle on the old map and still lose the war.

Your next competitor has no capital

There is a second-order effect that raises the stakes on all of it. The cost of starting a company has collapsed — the seed capital that used to take millions now takes thousands. You can argue about the exact multiple; the direction isn't in dispute. Your next serious competitor is probably not the incumbent you have been benchmarking against for years. It is two people with almost no capital who rebuilt your core workflow natively on AI — without the legacy systems, the headcount, or the organizational antibodies you are busy fighting.

Incumbency used to be a moat. Increasingly it is a liability, because the things protecting you — your scale, your processes, your installed base — are the same things slowing you down. The incumbent's edge today is not its size. It is whether it can move intelligence into its operations faster than a two-person team can rebuild those operations from zero. Most can't. That is the race, and almost no one is running it on purpose.

The window is shorter than you think

And it is arriving faster than any previous technology shift. A meaningful share of the planet is already using AI in some form, most of them without realizing it, and adoption is still accelerating. The industrial revolution took generations to reprice labor. This one is repricing knowledge work in quarters. There are even larger waves forming behind it, but that is a separate essay. For any executive reading this, the relevant point is narrower: the window to adapt is not a decade. It is the next few budget cycles.

Stop asking the bubble question

So when someone asks me whether AI is a bubble, my honest answer is that the chatbot might be a little overhyped, and it doesn't matter, because the chatbot was never the point. Pricing the revolution off the part you can see is a rounding error against the part you can't.

The better question — the one I would put to any COO, CFO, or CEO this year — is not "is this a bubble," and it is not "which model should we use." It is two questions. Where in our operations is human expertise the bottleneck? And what is actually stopping us from putting an agent there, accountably, and trusting it to act?

Answer those two honestly and you'll find the bubble debate was a distraction the whole time. The companies that win the next few years will not be the ones with the smartest model or the slickest demo. They will be the ones that did the boring, hard, organizational work of getting intelligence to do real work — while everyone else argued about whether the chart had peaked.

I know which group I would rather be in.