Imagine this scene. On Friday, a founder watches a factory quality manager spend two hours assembling the records for a rejected batch. By Monday, she brings back a tool that pulls the relevant documents into a single file. The quality manager likes it. Then he pulls up a night-shift batch and shows her what’s missing. The supervisor recorded a deviation in a shift log the tool couldn’t access. Even with the paperwork complete, the batch would still have to wait for an authorised reviewer to sign it off.
The founder built in a weekend what might once have taken several weeks. She came away with a clearer picture of where the factory was losing time. That finding shapes what she builds next, who she needs on the team and what she can credibly promise a customer.
Andrew Chen has asked which kinds of founders might thrive as AI makes it easier for individuals and small teams to build. There is a second question inside that one. What will entrepreneurs do differently when they can put an idea in front of a customer before they’ve assembled the company around it?
The first build can be an experiment
Consider a founder working on freight invoices. A finance director says her team spends too much time checking them. Automating invoice review sounds like a good place to start. With a rough tool and a month of sample invoices, the founder can sit alongside an analyst and see where the time actually goes. Suppose routine invoices clear in seconds. The work piles up around a small number of disputed surcharges, which send the analyst looking for a rate agreement, a delivery record or an email from the carrier.
A tool that prepares the evidence for a disputed charge could be more valuable than one that processes every invoice. In the sales conversation, the founder can then ask how much money the customer recovers on disputed charges, who owns that number and what would justify a budget.
AI makes it easier to build the broad product and the narrow experiment. The founder still has to decide which one is worth a week. It can be worth building something she expects to discard if it helps her decide what to pursue.
The bottleneck moves at its own pace
Imagine a team developing a sensor to detect wear in industrial pumps. AI can help it build software to review readings and compare signals across tests, giving the team a head start on the field trial. Once it is running, a different clock takes over. The pump has to run under different loads long enough for the team to establish what normal looks like. The maintenance team has to see whether an alert is early enough to act on. A false alarm can mean sending a technician to a remote site.
The founder could spend the next month adding features to the dashboard. Or she could spend it with the maintenance team working out which alert would actually change an inspection schedule. That work may give her a better sense of whether anyone will buy the sensor.
The same tension shows up in the factory. Now that the prototype has surfaced the sign-off problem, the harder step is getting the night shift to use the tool on a live batch. Watching how the reviewer handles it will help the founder decide what to build next.
The first hires can follow the evidence
A cheaper prototype gives the factory founder time to learn how the plant works before deciding who to hire. If customers need a tool that works across fragmented systems and preserves an audit trail, her first senior hire might be an engineer who has deployed software in industrial environments. If the plant needs to change who signs off on a batch, she may need someone who has led that change from inside a factory. The freight founder faces a similar choice. If the hardest part is interpreting exceptions buried in carrier contracts, an experienced freight operator could be her most valuable early hire.
At the idea stage, either of the factory’s needs might have sounded like “we need a technical co-founder”. Working through a live batch helps the founder see which skills she needs on the team and where a partner could help.
The company can take shape later
For investors, a polished demo still tells us something about how a founder thinks. A live trial can also clarify what the next round would buy. For the sensor company, that might mean enough field deployments to see whether the same signal predicts wear across different sites. That is a different investment proposition from funding a larger team to build out the product.
What counts as progress in deep tech?
AI can help a team get to its next milestone sooner. For investors, the question is what that milestone actually proves. A pump sensor that detects wear on a test bench shows that the signal is measurable under controlled conditions. A field trial can show whether it holds up across different loads and operating environments. A maintenance team using an alert to schedule an inspection shows that the signal arrived early enough to change a decision. Each step supports a different claim about the company.
At Entropia Capital, we ask founders what they can say with confidence today that they couldn’t say three months ago, and what the next round of capital will allow them to find out.
That brings us back to Monday. The factory founder walked in with a tool to save two hours of paperwork. She walked out knowing who controlled the release of the batch and what she needed to observe on the next shift. AI helped her get to that conversation sooner. Her next decision shows what she learned from it.


![Free Download] Factory Background Illustration (Horizontal) — Free ... Free Download] Factory Background Illustration (Horizontal) — Free ...](https://substackcdn.com/image/fetch/$s_!dnU2!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F1852f9ec-f02d-465c-b704-6221fa3e4ff3_1376x768.jpeg)
The hiring section made me think about how easily “we need a technical co-founder” can become a catch-all. Integrating older plant systems and changing who can approve a batch are both hard jobs, and they call for very different experience. I’d be curious how you decide which capability has to live inside the company and which a customer or partner can supply.
The pump example made me think about what a maintenance team does with an alert at 2 a.m. An accurate warning can still be hard to act on if the technician is hours away or the plant has no spare pump. I’d want the field trial to capture the decisions people make after an alert, alongside the quality of the signal itself.