Editorial: Hard tech needs customers who share the risk

A machine can perform beautifully in a demonstration and remain almost impossible to buy. The missing piece may be an installation window, a maintenance plan or somebody inside the customer organisation willing to own the trial. As attention turns towards hard tech, founders and the institutions supporting them should spend more time on this gap. Cheaper development is valuable. A workable first customer relationship is what allows that development to become a business.

Y Combinator’s 17 September episode, The State of Startups in 2026, describes a shift towards companies building in the physical world and smaller teams tackling ambitious problems. Those are observations from the accelerator’s own vantage point, not a census of entrepreneurship. The publisher’s description provides the starting point for this editorial. It does not establish that every hardware market has become easier to enter, or that the remaining barriers can be solved with software.

The distinction matters because physical products meet customers through operating systems that already exist. A warehouse has established shifts, loading areas and service commitments. A factory has machinery that cannot simply pause for an interesting experiment. A building manager has to coordinate access with tenants. The new supplier is asking the customer to accept disruption as well as a product. A persuasive slide explaining future efficiency does not decide who carries that disruption today.

Consider an illustrative inspection device that could reduce time spent checking industrial equipment. A founder might reasonably want access to a live site. The customer might reasonably want proof before providing it. Neither side is being irrational. The commercial task is to design a smaller agreement that makes learning possible: a defined inspection area, a named operator, a comparison with the current method and a process for stopping the trial. That agreement is part of product development.

This is where public support can be most useful. Norway’s innovation contracts programme, for example, explicitly centres committed collaboration between suppliers and pilot customers. The broader lesson is about participation, rather than copying the programme’s eligibility rules. A customer who provides operating knowledge, people and test access contributes something a grant cheque alone cannot supply. Support should make that contribution easier to organise and easier to evaluate.

For founders following startup and entrepreneurship channels, the practical challenge is to turn an exciting opportunity into a proposal a customer can act on. The proposal should explain the work required from each side. Who installs the device? Who trains the operator? What happens to equipment or data when the trial ends? These details are not administrative leftovers. They determine whether an interested customer can obtain internal approval and protect time for the project.

There is an equally important obligation on the supplier. A pilot should test a product that could serve a wider market. A young company can exhaust itself building an elaborate exception for its first prestigious customer. The relationship looks like validation while the business becomes a bespoke engineering service. Such work may be commercially sensible, but it should be recognised and priced as such. Calling every custom request a product milestone conceals the choice.

The first customer should therefore be selected partly for representativeness. Does the operating environment resemble those of future buyers? Are the requirements common, or unusually specific to one facility? Can results be described without breaching legitimate confidentiality? A smaller, less famous customer may be a better learning partner than a household name that offers little access to the people doing the work. Prestige is not a substitute for usable evidence.

The same discipline should shape the commercial outcome. Before a trial starts, the parties need to understand what a successful result would enable. It might lead to a paid deployment, another defined test or a decision by a procurement committee. There is no need to pretend that every pilot can guarantee a sale. There is every reason to name the next decision and its owner. Otherwise success can leave the startup stranded in a polite conversation.

Readers of leadership and business growth discussions will recognise a familiar organisational problem here: shared enthusiasm does not create shared accountability. A customer champion may love the technology but lack authority over budget or operations. The founder needs to know who can authorise the next step and what information that person requires. This is not an invitation to bypass the champion. It is a way to help the champion build a credible internal case.

The strongest objection is that early companies cannot afford this much process. They need speed. But a clear trial need not mean a thick document or months of meetings. It can mean a short agreement that prevents expensive misunderstandings. Speed is lost when a team arrives at a site without access, when nobody owns the evaluation or when success turns out to mean different things to different people. Preparation protects scarce engineering time.

For the customer, participation should also be proportionate. A trial should not consume an operational team’s entire improvement budget merely because the supplier is young. The supplier should propose a bounded commitment and make the learning useful to both sides. It should be possible for a customer to decline or stop without turning the relationship into a reputational dispute. A fair exit makes an initial yes easier to give.

The discussion in technology and AI channels often moves quickly from new capability to large economic claims. Between the two sits deployment: the ordinary work of integrating a tool into a real business. Hard tech deserves ambitious investors and skilled engineers. It also deserves purchasing arrangements that let useful products prove themselves. If we want more companies to build in the physical world, we should help more customers become competent partners in that work.

▶

Source episode: Y Combinator, 17 September 2026. Episode references are limited to the publisher’s description; the full recording was not reviewed.