The cost of creating a product has fallen dramatically. The cost of committing to the wrong product has not.
A founder can now describe an idea to an AI system and receive something remarkably convincing in return. It might be a website, an application, a customer dashboard, a booking system or a complete set of product screens. What once required several specialists and considerable expenditure may now be produced by one person, sometimes in days. The results can look professional. The software may function. The founder can demonstrate something tangible rather than explain an idea that exists only in their head. That is a genuine advance, but something important has changed alongside it. It has become much easier to produce something that looks like a business before establishing whether the business behind it makes commercial sense. And that distinction is becoming increasingly expensive to overlook.
The investor who doesn't want founders to hire anyone
In a recent article, product strategist Dan Maccarone described a conversation with a venture capitalist who reportedly would not invest in a founder who hired someone to build or design their product. The reasoning was straightforward: with today's AI tools, founders should be capable of doing much of this work themselves. There is a reasonable argument buried in that position. Early-stage businesses have often spent heavily on development teams, agencies, technology and professional services before they had sufficient evidence that the resulting product was required. Being able to produce an initial version inexpensively is an advantage. It can preserve money and allow a founder to learn without making a large commitment.
But the investor's argument becomes considerably weaker when it moves from reducing unnecessary expenditure to assuming that the ability to produce something establishes the competence to judge it. Knowing how to make an application is not the same as knowing whether it is safe, reliable, accessible or suitable for its intended users. And there is another question beyond those.
Even if the application is excellent, should the founder be building it at all?
Three questions that are easily confused
Consider a founder developing software for independent tradespeople. Using AI, they create a polished platform for receiving enquiries, producing quotations, booking appointments and managing payments. The screens are attractive and the features operate as expected. The founder may reasonably regard this as substantial progress. But three separate questions remain.
First: Can it be built? The founder has demonstrated that a working version can be produced. Second: Has it been built properly?
That requires different evidence. Does it handle errors, protect customer information, remain reliable, and support the practical circumstances in which people will use it?
Third: Does the business need this product in the form proposed? That is a commercial question. Perhaps the intended customers already have adequate software. Perhaps they have little willingness to pay for another subscription. Perhaps their actual difficulty is finding profitable work rather than managing appointments. Perhaps the founder can reach these customers but cannot acquire them economically at the price they are prepared to pay. None of these possibilities is resolved by making the application better. A technically competent product can sit inside a commercially weak business. The first question is about production. The second is about professional competence. The third is about commercial reality. AI has made remarkable advances against the first and can contribute meaningfully to the second.
It has not made those three questions interchangeable.
The appearance of progress
There is a particular danger in the speed and apparent completeness of modern AI-generated work. Previously, an undeveloped proposition was often visibly undeveloped. A founder had a sketch, a document or perhaps a basic prototype. The distance between the idea and an operating business was difficult to ignore. Today, the same founder can produce a convincing presentation, a working website, a financial forecast and a product demonstration before speaking to a single independent customer. Everything can appear remarkably advanced. Yet the underlying commercial assumptions may be almost entirely unchanged. The professional appearance of the output can obscure how little has been established about what customers will buy, how they will be reached, what delivery will cost, or how the business will generate sufficient cash.
This is not a theoretical concern about technology. Research on AI-assisted professional work has demonstrated that its benefits vary according to the task. In a study involving 758 consultants, researchers found substantial improvements when AI was applied to tasks within its capabilities, but worse results on a task outside those capabilities. The tools were productive. They were not uniformly reliable. And the people using them did not always have an obvious way to distinguish the two situations. The commercial equivalent can be more consequential. A product can appear finished while the assumptions justifying its existence remain unexamined.
The hidden cost of producing things cheaply
There is an understandable tendency to describe AI-generated work as inexpensive. A founder may spend very little creating an initial version of something that previously required thousands of pounds. But the cost of initial production is only one part of the decision. A functioning prototype can quickly become the basis for other commitments. The founder begins paying for hosting, specialist integrations and software subscriptions. They commission improvements. They make promises to prospective customers. They establish supplier relationships around the chosen technology. Months of work can accumulate around an operating structure that was never commercially established. The original prototype may have cost almost nothing. The commitments built around it may not. And there is another cost that often receives less attention: the founder's time.
A person who spends three months building something with AI has still spent three months. Those months could have provided useful commercial learning. Equally, they could have been devoted to producing more of something customers were never likely to buy. Whether the founder wrote the code personally, employed a developer or used an AI system does not change the commercial consequence. Low production cost does not automatically mean low decision exposure. The crucial distinction is between using inexpensive tools to learn and allowing inexpensive output to justify increasingly expensive commitments.
Expertise is not the same as headcount
The response to these concerns should not be to insist that every founder needs a large professional team. That would ignore one of the most valuable changes AI has created. Some founders genuinely possess the technical, industry and commercial knowledge necessary to handle substantial parts of their work independently. Others can produce useful early versions without immediately employing people they cannot afford. Research into AI-assisted customer service has also found that less experienced workers can benefit substantially from AI assistance. It would be inaccurate to claim that expertise can never be developed or supported through these tools. But the ability to obtain an answer is not necessarily the ability to recognise where an answer is inadequate.
A founder who understands their industry may have little experience of software reliability, payment security, customer accessibility or contractual obligations. A skilled developer may know how to deliver an excellent product while having limited knowledge of its intended market. A professional designer may improve customer comprehension and usability without being able to establish whether the underlying offer is commercially worthwhile. Each form of expertise can address a different part of the problem. And expertise itself is not infallible. An experienced professional can still work from incorrect assumptions, misunderstand a market or confidently endorse something customers do not value. Hiring an expert does not establish commercial evidence any more than using AI does.
The relevant question is whether the founder can identify what knowledge the decision requires, what is already understood, what remains uncertain, and what evidence is needed before the business becomes committed. Expertise may be available within the founder, through another person, through specialist review, or through a combination of people and technology. Its value comes from addressing a genuine gap, not from increasing the number of people involved.
The decision has not disappeared
There is a temptation to imagine that the advancing capability of AI will eventually remove much of the difficulty involved in creating businesses. And some difficulties undoubtedly will become smaller. Technology can reduce the cost of exploration. It can make specialised capabilities more accessible. It can help a founder see possibilities that would previously have been beyond their practical reach. But business decisions still create consequences. Someone must decide which customers matter, which commercial assumptions deserve testing, which requirements are essential, which commitments are affordable, and what evidence would justify changing direction. AI can contribute to these decisions. Experts can inform them. Customers can challenge the assumptions behind them. None of those inputs automatically establishes that the resulting decision is commercially justified.
A founder also cannot delegate away the consequences. If the software is unsuitable, the contract unaffordable, the delivery uneconomic or the market absent, it is the founder's business that carries the exposure. This is why the debate about whether founders should build everything themselves is too narrow. One founder may create an excellent business with very little outside help. Another may require several specialist disciplines. A third may discover that the proposed business does not justify further commitment at all. The correct configuration cannot be established from a blanket rule about how many people should be involved. It depends on what the business actually needs and what is already known.
The question worth asking
The ability to make things is becoming more widely available. That is a significant opportunity for people who previously lacked the money or technical resources to turn an idea into something tangible. But greater production capability also makes it easier to move from an untested assumption to a convincing finished product without confronting the commercial questions in between. The danger is not that AI produces poor work; sometimes it produces very good work. The more uncomfortable possibility is that it produces exactly what the founder requested, even when the request itself was commercially wrong. A founder does not need to master every discipline involved in building a business. Nor does a founder need to employ specialists simply to demonstrate that expertise is being taken seriously.
But before allowing a compelling output to become an expensive commitment, there is a question that neither technical sophistication nor professional presentation can answer on its own:
What do I actually know that makes this the right thing to build, and what am I committing before I have established it?
Sources
Dan Maccarone, “A Founder With a Tool Is Not a Founder With a Clue”, 21 October 2026.
