Two Types of AI: Why Most Companies Lie to Themselves

Most companies claiming “AI-first” are either doing commodity work while claiming innovation, or attempting frontier breakthroughs they’re unequipped for. Understanding which determines everything: structure, funding, hiring, timelines.

The founder leans forward across the conference table. “We’re building an AI-first company,” he says. “Our platform uses machine learning to transform how enterprises handle customer data.”

Fifteen minutes later, the truth emerges. The “AI” is GPT-4 through an API. The machine learning is a standard classifier from an open-source library. There is no proprietary algorithm, no research team. The company is doing something useful, but nothing that a dozen competitors couldn’t replicate in a quarter.

This founder genuinely believes he is building an AI company. By today’s diluted standards, maybe he is. But by any meaningful strategic definition, he is wrong. And that wrongness will cost him — in misallocated capital, poor hiring, unrealistic timelines, and a business model with no moat.

The term “AI company” has been so debased it no longer conveys useful information. To build a durable business, leaders need a framework that cuts through the noise. That framework begins with one critical distinction: Commodity AI versus Frontier AI.

Commodity AI

Commodity AI is the application of ready-made, off-the-shelf AI tools to deliver business value. Using OpenAI’s API. Deploying established computer vision models. Implementing standard recommendation algorithms.

Commodity AI has three characteristics:

• Low risk. The technology is proven. The path from problem to solution has been demonstrated repeatedly. Technical uncertainty has been resolved.
• Fast and cheap. You need competent engineers, not AI researchers. Implementation takes weeks or months, not years.
• Value from business expertise. You win by understanding your domain, knowing where these tools create customer value, and executing efficiently.

Frontier AI

Frontier AI is fundamentally different. It means pushing beyond current state-of-the-art capabilities. Confronting core limitations that have stumped researchers. Achieving genuine breakthroughs in what AI can do. Creating capabilities that do not currently exist.

The distinction is not about sophistication. A commodity application might look impressive. A frontier system might look crude early on. The distinction is about where value comes from. With commodity AI, the technology is available to everyone — value comes from business execution. With frontier AI, value comes from the technological breakthrough itself that others cannot replicate.

The Two Types of Hype

The first type comes from companies misrepresenting commodity work as frontier innovation. They talk about “proprietary algorithms” and “cutting-edge machine learning” while wrapping API calls in business logic. The danger is the mismatch between positioning and reality. They raise capital on frontier promises, then have to explain why competitors replicated their “breakthrough” in six months.

The second type is more dangerous. These companies vastly overstate what current AI can do. They are naive to fundamental limitations of the state-of-the-art, dismissing them as minor engineering challenges rather than recognizing them as hard boundaries of what is currently possible. They present timelines that assume away years of unsolved research. They staff with capable engineers rather than researchers who have actually pushed past these boundaries before.

These companies do not understand they are attempting something that has defeated better-resourced teams with deeper expertise. Their failure is not a possibility — it is inevitable. The only question is how much capital and time will be burned before reality becomes undeniable.

What Happens When You Get It Wrong

If you claim frontier but do commodity:

• You raise capital on wrong terms. Frontier investors expect high risk and long timelines. When your “frontier” work turns out to be wrapped API calls, they feel misled.
• You hire the wrong people. Frontier work requires PhDs with publication records. These people are expensive and will leave when they discover the work is commodity.
• You set impossible timelines. Commodity work has predictable schedules. Frontier breakthroughs do not.

If you do frontier but position as commodity:

• You underprice your value. Frontier capabilities command premium prices because no one else can provide them.
• You attract wrong scrutiny. Commodity businesses are evaluated on execution and unit economics. Frontier businesses are evaluated on technical credibility and defensibility of breakthrough.

Both Paths Are Legitimate

Most companies are building commodity AI, and that is not only okay — it is the right choice.

Commodity AI is not second-class. It is how most business value from AI will be created. Research labs push boundaries, then thousands of companies apply those capabilities to create customer value. Winners in commodity AI are those with the best domain understanding, strongest customer relationships, most efficient operations, and clearest thinking about where AI creates genuine value.

If you are building commodity AI, embrace it. Stop using frontier language. Talk about your domain expertise, customer workflow understanding, and business model. These are the real sources of value.

If commodity AI is more common than people admit, frontier AI is much rarer and much harder than people realize. Real frontier work is an exercise in humility. You are trying to do something the best researchers in the world, at institutions with enormous resources, have not figured out how to do.

The Strategic Question

For most companies, the answer is commodity. Become sophisticated and aggressive in deploying proven capabilities. Structure for execution speed, customer intimacy, and operational efficiency.

For a select few, the answer is frontier. You have identified a genuine gap in current capabilities, you have or can assemble the necessary ingredients, and you accept high risk and uncertainty in exchange for potential to create an unassailable position.

Both paths are legitimate. Both can build valuable businesses. The error is not in choosing one or the other. The error is choosing one while pretending to be on the other.

The first step toward successful AI strategy is being honest about which path you are actually on. Most companies are lying to themselves about this.

Stop lying. Choose deliberately. Execute accordingly.