At the beginning of 2026, I dropped out of Penn with my cofounders with a thesis that AI would change the industry of manufacturing entirely. Starry-eyed and funded by YC, we set out with dreams of changing a pillar of the global economy.
Prior to our application, I had spent months tossing around potential ideas with Daichi (my roommate and eventual cofounder), but none of them really gained any traction. We had briefly tried pursuing data annotation for AI CAD in August of 2025, but quickly concluded we weren’t the right founders to really tackle such a technically challenging problem. Combined with a lack of passion/conviction for the idea, we decided to keep waiting until something else came along.
After a few talks with potential investors and relatively low interest, we decided to tackle problem validation by approaching and talking to as many friends and family as we could. This started with my aunt’s small trading company. At the time, they were a small team relying on archaic spreadsheets to process millions in volume. We had relatively little experience with industry, but it was quite easy to see the inefficiencies present throughout the process. Based on their feedback and daily workflows, we mocked up an agentic ERP and submitted it to YC, eager to set out and conquer the problem. The underlying thesis was theoretically sound: use AI to solve manual work, and profit!
We ended up being accepted, and I still remember barely being able to sleep the night after our interview, staring at the phone whilst waiting for a call. A few days later, we said goodbye to our friends, held a few going-away parties, and set out for Silicon Valley.
We got in early and had two months before YC started, so during the early stages of problem validation, the team flew to Asia to build the product out of my aunt’s office. The goal of a startup is always to create the best feedback loop possible. The more information and iteration flowing throughout the product, the faster you can grow and refine.
It was great in theory. However, we messed up in two ways:
We overbuilt every feature. As ambitious young founders, it was natural to want to build using the best technology. Instead of talking to our customers more, we spent weeks investing in heavily overengineered code that was barely better than the incumbents. As a founder, the group that you should be talking to the most after your cofounders is your customers. The goal is to build the best product for their needs, and not the version of them you have in your head.
We never bothered to truly understand the market. We were working with a trading company, yet wanted to target the manufacturing industry. We assumed that working with some manufacturing knowledge from my time in automotive automation was enough, and that things would just scale. We never bothered to figure out if other people had the same problems, and wasted several months building a product with no real input. Our expectations didn’t meet reality, and the product wasn’t something that people ended up resonating with.
Soon after, our time was up and we moved to SF for YC. While we kept working on the ERP for a month after the batch started, we soon decided to pivot. At the time, we believed that the problem we were facing was that people didn’t want another ERP to migrate to. Thus, we decided to build AI agents that just did the work that existing employees did, instead of tools to help them do so.
I want to say that we learned from our mistakes, but we didn’t. Our insights came from trying to self-validate the direction we were heading in, and once again, we began building a product without the customer. We scheduled many meetings with employees who were high up in manufacturing, and combined with deep research using Claude, were satisfied enough to proceed. In reality, we didn’t really talk with anyone who would actually be using our product on a daily basis. We weren’t out on the factory floor watching workers complete their tasks or navigate the ERP. Out of all the mistakes we made, this was probably the biggest one.
We did eventually get our first customer, but ran into more problems with tackling a legacy industry like manufacturing. The ones presented below were a consensus from talks with other companies (in both our and future batches) that were tackling a similar problem to ours.
Firstly, data sanitation was the main problem. The data that was present in these ERPs was often fragmented, and a good portion of each company’s information would flow through inaccessible spreadsheets/private messages. The amount of work that it took to clean these processes up and gather all of the prior data made scaling impossible.
Additionally, middle-aged operators were often turned off by new technology, no matter the form. They preferred what they had used for decades already, and it was extremely hard to have any real adoption amongst the primary workforce driving these businesses. Talking to business owners, we were often the second or third initiative that they had tried. While I believe this is not the case for all factories, it certainly was for the ones that we interacted with.
There were a million other minor issues that ended up stacking against us. Together, it made growing extremely hard, and we concluded that we were never going to find product-market fit in the direction we were going in.
Around this time, our batch had a talk with one of YC’s founders, Paul Graham. I had always been a fan of his essays, and am constantly reminded of this quote:
“since your intuitions about other people’s needs are usually a crap signal, and your own needs are an especially valuable one, you should usually listen to the second signal; you should make something you and your friends want.”
Manufacturing wasn’t an industry we had an exceptional amount of passion for; it was just one that looked ripe with opportunity.
We spent a few weeks after YC taking a break and exploring new ideas. Working unreasonable hours during the batch taught us to pursue a more routine schedule, and now the team is back and building something that actually matters to us personally.