Why AI Companies Should Sell the Shovel, Not Claim the Gold
1. The Basic Business Principle
There is an old business lesson that is particularly relevant to AI: do not compete with your customers.
When Adobe built Photoshop and Illustrator, it did not need to become the artist. When Corel built CorelDRAW, it did not need to claim ownership of the illustrations produced with it. These companies built increasingly powerful tools and allowed their customers to create the masterpieces.
The masterpiece belonged to the artist. The software company demonstrated the superiority of the tool.
That creates a remarkably healthy relationship. The better the customer becomes, the more valuable the tool becomes. A great designer using excellent software is not a threat to the software company. The customer’s success is the company’s success.
2. The AI Temptation
AI creates a new temptation. When an AI system helps someone write a paper, develop software, design an image or investigate a mathematical problem, it becomes easy to blur the distinction between what the human achieved with the tool and what the tool itself achieved.
That distinction becomes critical when a company starts making extraordinary claims about its AI. There is a huge difference between saying, “Our AI helped a mathematician solve a difficult problem,” and saying, “Our AI solved a difficult mathematical problem.” The first is a claim about a powerful tool. The second is a claim about autonomous intelligence.
Once a company makes the second kind of claim, it is no longer merely selling software. It is selling a story about what the software itself is capable of.
3. Sell the Shovel, Not the Gold
The gold-rush analogy is almost perfect for AI. During a gold rush, you can search for gold yourself, or you can sell tools to thousands of people searching for it. The second model has an interesting advantage: you don’t need to know where the gold is. You only need to build a better shovel.
That principle could be extremely powerful for AI companies. An AI company does not need to prove that its AI personally discovered every theorem, wrote every great program or made every scientific breakthrough. It can build tools that allow mathematicians, scientists, engineers, programmers and artists to achieve things they could not achieve as easily before.
If those customers produce extraordinary results, the AI company benefits anyway. Better tools create better customers; better customers create better results; better results create demand for better tools. There is no need to claim ownership of the gold.
4. Never Compete With Your Customers
Imagine an AI company providing a mathematician with an extraordinary research assistant. The mathematician explores thousands of possibilities, finds an unexpected structure, develops a proof and eventually publishes a breakthrough.
Who should get the credit? The mathematician.
The AI company can legitimately say, “Our system helped make this possible.” That is already an extraordinary achievement for the technology. There is little to gain by saying, “Actually, the AI solved it. The mathematician merely supervised it.”
The customer is no longer being treated as the beneficiary of the technology. The customer is being turned into a competitor for intellectual credit. That is a strange business strategy.
A technology company should want its customers to look brilliant when they use its product.
5. The Navier–Stokes Warning
The recent controversy surrounding OpenAI’s claimed Navier–Stokes solution illustrates why this distinction matters. OpenAI presented the result as an extraordinary demonstration of its internal AI’s mathematical capability. The system reportedly worked through the problem using a huge number of agents and subsequently produced a formalized proof.
That creates a very high bar. If a mathematician says, “I have an idea about Navier–Stokes,” the world expects the idea to be checked. If a company says, “Our AI solved Navier–Stokes,” the statement becomes a demonstration of the AI’s intelligence itself.
Now suppose independent mathematicians identify a fundamental problem in the argument. The issue is no longer merely that a mathematical paper needs correction. The company’s capability claim is also damaged.
That is the danger of claiming the gold instead of selling the shovel.
6. Mathematics Is an Unforgiving Customer
Mathematics is perhaps the worst place to overclaim. It is objective. A theorem does not become 99% true because the remaining 1% looks plausible. A proof does not become valid because the equations are sophisticated or because a computer has verified part of them.
One invalid step can invalidate the entire proof.
That makes mathematics both exceptionally good and exceptionally dangerous for AI. It is good because AI can exploit something mathematics provides in abundance: objective feedback. The system can calculate, test, search, formalize and verify.
But mathematics is dangerous because eventually the system has nowhere to hide. A beautifully written explanation does not matter if the conclusion does not follow. A million lines of symbolic manipulation do not matter if one necessary assumption is missing. A green verification mark does not establish that the machine proved the intended theorem if the formalized statement differs from the original claim.
Mathematics is therefore a particularly powerful stress test for AI. It rewards genuine correctness and exposes counterfeit correctness.
7. When AI Solves “How to Solve”
This leads to a deeper question. Perhaps AI is not always solving mathematics in the same sense that a mathematician solves mathematics. Perhaps it is increasingly becoming extraordinarily good at solving the problem of how to solve mathematics.
It can search enormous spaces of possibilities, generate lemmas, rewrite equations, try alternative approaches, discover useful representations, call other agents, write computer code and attempt formal verification. That is tremendously powerful. But it is a different claim from understanding the mathematical problem and constructing a valid proof from that understanding.
There is a difference between “I understand this mathematical problem and have proved the theorem” and “I found a sequence of operations that produced something that appears to solve the problem and passes the available checks.” The second capability could become extraordinary without necessarily implying the first.
This is one reason AI-generated mathematics can sometimes look like word salad mixed with symbol garbage: sophisticated mathematical language and notation can create an impression of understanding that is much easier to generate than genuine mathematical validity.
The danger becomes greater as systems become better at finding ways through verification systems. The question then changes from “Can AI produce a proof?” to “Can AI reliably prove that it has proved the thing we actually asked it to prove?” That is a much harder problem.
This question also connects with the broader distinction explored in Why Inventing Is Not the Same as Understanding: producing an impressive result and understanding why the result is correct are not necessarily the same capability.
8. The Danger of Falling in Love With Your Creation
There is also a human problem here. Creators naturally become fascinated by what their creations can do. When an AI system suddenly produces something astonishing, the temptation is to extrapolate: it writes beautifully, then it reasons, then it solves difficult problems, then it discovers mathematics, then it must understand mathematics, and perhaps it is approaching AGI.
But every arrow in that chain is another claim. Capability is not the same thing as understanding. Output quality is not the same thing as reliability. A successful demonstration is not automatically a general capability.
The discipline required from an AI company is therefore similar to the discipline required from a scientist: be impressed by the result, but be even more impressed by the verification.
9. Underpromise, Overdeliver
There is a simple management principle that explains why this matters commercially. Suppose one manager promises a 20% improvement and delivers 50%. Everyone celebrates. Another promises 80% and delivers 60%. Everyone asks what went wrong.
The second manager actually delivered more. But the first delivered 250% of the promise, while the second delivered only 75%.
The lesson is simple: underpromise and overdeliver. The reverse—overpromise and underdeliver—is dangerous.
This is particularly important for AI because expectations have become part of the industry’s economic value. If a company sells an AI as an extraordinarily useful tool, occasional failures are expected. If it sells the same AI as evidence that autonomous scientific intelligence has arrived, every serious failure becomes evidence against the larger claim.
The bigger the promise, the bigger the penalty.
10. Let the Customer Win
The most successful AI companies may eventually discover that they do not need to win every intellectual competition themselves. Let the mathematician win the theorem, the scientist win the discovery, the programmer win the software, the artist win the masterpiece and the entrepreneur build the company.
The AI company can still win. It wins by making all of them more capable.
That is not a smaller ambition. It may actually be a much larger one. Imagine a world in which millions of people can do work that previously required exceptional expertise because AI gives them an extraordinary intellectual toolkit. The AI company does not need to own those achievements. It created the multiplier.
11. The Real Competitive Advantage
There is a profound strategic difference between replacing the customer and empowering the customer.
A company that tries to prove its AI is smarter than everyone eventually creates a difficult question: If the AI can do everything, why do I need the customer?
A company that builds tools that make customers dramatically better creates the opposite question: How did anyone work without this tool?
That is a much stronger position. Adobe did not need to become the world’s greatest artist. Corel did not need to become the world’s greatest designer. Their tools became valuable precisely because other people could become great with them.
AI can follow the same principle.
12. Open Advice to OpenAI
Sell the shovel. Don’t claim the gold.
OpenAI has built one of the most powerful intellectual tools in history. It does not need to prove that the tool itself is the mathematician, scientist, programmer or artist. The greater achievement may be to make millions of those people dramatically more capable.
Let the mathematician own the theorem. Let the scientist own the discovery. Let the programmer own the software. Let the artist own the masterpiece.
OpenAI should compete with other AI companies for better tools—not with its customers for intellectual achievement.
The temptation to fall in love with your own creation is understandable. But business is business. A tool can survive a mistake. A grand claim can turn a mistake into a crisis.
Build the best shovel in the world. Let your customers find the gold.
Don’t compete with your customers for the achievement. Make yourself indispensable to their achievement.