On the first of May I walked out of my university research lab for the last time. The plan I was walking away from was a doctorate — about six more years of it. The plan I was walking toward was nothing at all. I had no idea what I wanted to build. I just had a suspicion that with AI as a working partner, the things that used to take six years might take three months.
That suspicion turned out to be about half right, in the way suspicions usually are.
Fifteen weeks later I’m running an EV battery startup that has measured six cars, visited four customers in their own driveways, and collected exactly zero won in revenue. This is the honest version of those fifteen weeks — three ideas, two rejections, one very quiet day at a dealer lot, and the reason I’m still going.
My first idea had a total market of about seventy thousand dollars
Not seventy thousand per customer. Seventy thousand dollars a year, for the entire market. I just hadn’t checked.
The idea came from the only thing I was genuinely expert in: I’m the electrical lead on a student-built Formula Student EV car, and every year our team relearns the same painful lessons because the previous lineup graduated and took their knowledge with them. So: build the training material. An electrical systems curriculum for student car teams. I knew the pain intimately because I was the pain.
I put it into two funding applications, deadlines a week apart in May. Both said no, and they were right to. Somewhere in the middle of writing the second one I actually added up how many student teams exist and what any of them could plausibly pay. The number was so small I laughed at my own spreadsheet.
That’s lesson one, and it cost me three weeks: I had validated that the problem was real and never once checked whether the market was. Those are completely different questions and I had confidently answered only the easy one.
Three weeks of exams that changed what I thought the bottleneck was
June was exam period. Three weeks on paper, and because I’d spent May failing at a business, about two weeks of actual studying. I ran the whole thing through AI — not to get answers, but to be walked through the parts I didn’t understand until I did, at whatever hour I happened to be awake.
I finished first in two of my major courses that term. I’m not telling you that to brag about the grade. I’m telling you because of the input: two weeks. And the conclusion I drew from it has quietly organized everything I’ve done since.
Knowledge isn’t the scarce thing anymore. Deciding what’s worth building is. I could learn almost any technical subject faster than I ever could have before, which meant the constraint on my life had moved. It wasn’t “what do I know how to do.” It was “what should I point this at.” That is a much harder question, and nobody can outsource it to a model.
A room where nobody could tell me if my idea was good
In early July I spent three days at a startup camp with idea number two: a charging setup for company fleets, using a factory’s leftover electrical capacity to charge its own electric trucks overnight. I still think there’s something there.
I explained it for three days. And at the end of three days, not one person in that room could tell me whether it was a good idea or a bad one.
That is a genuinely bad thing to discover as a founder. Feedback is how you steer, and I had none. I remember sitting there pretty deflated about it. Then it turned over in my head and landed the other way up: if nobody in this room can evaluate it, that isn’t only a problem. That’s also a description of a moat.
In a world where anyone can learn anything quickly, the advantage that lasts isn’t knowing something. It’s the part you have to physically touch to understand. You can read every article ever written about battery packs and still not know what a real one does at eight in the morning in a stranger’s driveway, with a bad connector and a customer watching you work.
So I went looking for a growing market where the answers live in hardware. Electric vehicles. Then narrower: used electric vehicles — where the single most expensive part of the car is the one nobody can see. The US Department of Energy expects an EV pack to still hold at least 70% of its original capacity at the end of its life in the car, which sounds reassuring right up until you try to find out where on that scale one specific used car actually sits. Idea number three. That was July 9th.
The week I stopped thinking and went outside
Eight days later I was standing in a used car complex in Seoul with eight printed battery reports, trying to sell a service to dealers. I handed out all eight and sold nothing, and it took me another two weeks to understand that the reason wasn’t my age or my credentials — it was the incentive structure of the place. That day is still the most useful thing that has happened to this company.
Three days after that I rented a Kia Niro EV with 204,000 km on the odometer, plugged into the OBD port, and got my first real measurement. The car reported 97% battery health. Cross-checked against its own energy counters, I got closer to 89%.
That gap is the entire reason this company exists. Two months of thinking produced two dead ideas. Two weeks of going outside and measuring things produced a business. I don’t think that ratio is a coincidence, and I don’t intend to forget it.
Where my EV battery startup actually is, fifteen weeks in
Here’s the scoreboard, without the polish that usually gets applied to these things.
Six cars measured. Four real customers visited, in their driveways, on their schedules. Nine people have contacted me asking for this. Zero of them have paid.
Every visit so far has been free, because I’d rather have the data and the honest feedback than a small amount of money and no idea whether the thing works. But I’m not going to dress that up: a company with no revenue is a hobby with paperwork, and getting the first person to actually pay is the only real problem I have. Everything else is a distraction I’m choosing not to be distracted by.
The other half of these fifteen weeks was learning to build the thing itself. I got the race car’s CAN bus talking in a single day without knowing C — a job that had defeated a much stronger programmer on our team the year before — and the rest of that firmware took about two weeks. The diagnostic app came together the same way: fast, and then very slowly, because the last stretch is always where you find out what you assumed.
I won a startup competition somewhere in there. It was a nice afternoon. It did not make anyone pay me.
Three things I’d tell myself on May 1st
If you’re a student sitting on the same fence I was sitting on, these are the only three things I’ve learned that I’d actually defend:
- Size the market before you fall in love with the problem. A real pain with no market is a hobby. I spent three weeks proving something I already knew and zero minutes on the question that would have killed it in an afternoon.
- Go where the answer requires hands. Anything you can learn by reading, everyone else can learn by reading too, and faster every year. Whatever you have to touch, drive, open, or measure is where the durable advantage is hiding.
- Get outside sooner than feels reasonable. My worst day — eight reports, no sales — taught me more than my best week of planning. You cannot think your way to a customer.
Why I’m writing this down now
I’ve written the dream down before so I can’t quietly delete it later: I want to spend weekends at a circuit doing time attack in cars I love without checking my bank balance first, and I want to show up on graduation day in a BMW M2 that belongs to me. That’s still the plan. I’m aware of exactly how far away it is.
Most founder stories get told backwards, after it works, when every wrong turn can be narrated as a clever setup for the ending. I’d rather write mine while it’s still genuinely uncertain — while the revenue is zero, and the next measurement might contradict the last one, and I don’t know yet whether any of this becomes a company.
If it works, this record will be worth more than a tidy retelling. If it doesn’t, somebody gets a free and accurate map of exactly where the holes are. I’m fine with either — an EV battery startup that fails in public is still more useful than one that never showed its work. What I’m not willing to do is show you only the parts that went well.
The next car is already booked.
This is the origin story behind Battery Lab, where I measure real battery health on used EVs and publish what I find — including the days the numbers don’t agree with each other.