The Most General Tool Ever Built Sells One Specific Thing

A large language model is the most general-purpose technology I’ve ever used.

The initial reaction upon first interacting with one is universal: this can do anything. ChatGPT democratized that realization. A single text box handled everything from drafting poetry and debugging functions to planning itineraries or explaining quantum mechanics. It presented an interface that seemed capable of tackling whatever you threw at it.

Which makes the current market reality so counterintuitive: if the core capability is truly that universal, why are the products generating the most revenue so narrow?

When I look at what genuinely works as a sustainable business (tools like Claude Code and GitHub Copilot), product-market fit didn’t arise from generality. It came from pointing that raw intelligence engine at one specific, painful, and expensive problem: writing and maintaining software.

This is the recurring paradox: the most general technology ever invented finds product-market fit by getting specific.

Generality is not a business model

A blank text box capable of everything is, paradoxically, difficult to monetize at high margins.

The value it provides to any individual user is real, but it remains diffuse: scattered across a hundred occasional tasks, none of which carry an explicit business budget. ChatGPT is an extraordinary consumer product and a daily habit for millions. Yet there remains a substantial gap between “this is useful for random inquiries” and “we allocate a dedicated budget line for this exact function.” Commercial revenue lives on the far side of that line.

Product-market fit is rarely about what a technology could theoretically do. It is about identifying a problem painful and frequent enough that buyers are already looking for solutions. Generality represents supply. Demand is an entirely different matter: customers do not pay for raw potential; they pay for finished, reliable outcomes.

Why software engineering won first

Through this lens, it is obvious why software engineering became the first domain where agents gained commercial traction. Coding possesses an unusual combination of attributes:

• It is verifiable. Code either compiles or fails. Test suites either pass or break. The model receives unambiguous feedback on whether its output is correct, and so does the buyer. Without that automated feedback loop, trust is fragile, and nobody pays significant recurring revenue for software they cannot trust.
• It is expensive. Senior engineering payroll is high, meaning even a 10% gain in output translates into quantifiable dollar value. The ROI calculation is immediate.
• It is continuous. Software engineers encounter these hurdles all day, every single day. That relentless frequency transforms a pleasant convenience into indispensable infrastructure.
• It is already budgeted. Companies already maintain line items for IDEs, CI/CD runners, and developer tooling. You are not asking procurement to invent a new spending category; you are upgrading an existing one.
• The output is the final deliverable. What the model outputs (source code) is the asset. It does not require manual transcription or creative reinterpretation to create business value.

Generality allowed the model to comprehend syntax and logic in the first place. But the economic leverage came entirely from the surrounding commercial traits.

Paying without hesitation

The clearest signal of product-market fit is customer behavior under renewal.

When I pay for Claude Code, I do not deliberate over the subscription cost. Not because it is cheap, but because the alternative (my own development hours) is dramatically more costly. The return on investment is legible: I can point directly to features shipped and bugs resolved that would have consumed whole afternoons.

That legibility defines commercial success. General conversational chat is useful, but if pressed to quantify the exact dollar value saved by a general prompt session, most users hesitate. A dedicated coding agent never prompts that hesitation. When the return is so clear that the buyer stops calculating, product-market fit has arrived.

Lessons for product builders

When building on foundation models, the natural gravitational pull is toward building “an assistant that helps with everything.” The flexibility of the model tempts teams to avoid constraints.

Resist that impulse. The model’s breadth is precisely what tempts products into vagueness, and buyers rarely pay premium prices for vague tools.

Find the problem that is painful, costly, frequent, and already funded by existing budgets. Generality is merely your raw material; product-market fit comes from the sharp precision with which you apply it, and how easily buyers can verify its return.

The classic technology pattern

If this dynamic were unique to LLMs, it would merely be an industry quirk. In reality, it is the oldest adoption pattern in technology. Broad capabilities consistently prove their value through specific, narrow applications.

The early personal computer was a general computing platform that hobbyists loved while corporations remained skeptical. That skepticism evaporated not because of better marketing about general utility, but because of VisiCalc, the first electronic spreadsheet. Suddenly, financial accountants had an urgent, specific justification to purchase an Apple II. People bought general-purpose computers to execute a single high-value application. The killer app was hyper-specific.

When the laser was first invented, observers dismissed it as “a solution looking for a problem.” Coherent light itself had no commercial market. But barcode scanners, CD players, fiber-optic telecommunications, and laser eye surgery found immense markets. The underlying physics conquered industry one specific application at a time.

The internet provided a universal data substrate, yet nobody paid simply for “hypertext.” Customers paid for email, then search indexes, then e-commerce platforms. Amazon did not launch as the “everything store”; it started by selling books, and earned the right to expand outward.

The broad technology platform is the foundation; the narrow, repeatable, high-value workflow is where customers willingly exchange money. Coding agents are simply the latest iteration of this historic playbook.

Distribution broad, revenue focused

None of this implies general interfaces are unimportant. The open prompt interface of ChatGPT represents a staggering distribution vehicle and user acquisition engine. Over a long enough timeline, general platforms may gradually absorb specialized workflows.

However, where sustainable commercial revenue materializes today is far more focused than broad demonstrations suggest.

The most versatile technology ever created is establishing its economic foundation one concrete problem at a time. Far from a retreat in ambition, that is simply what real product-market fit has always required.