Michael Wellborn sat in his living room in Texas and watched Pacific Palisades burn.
He had grown up around that place. He watched a neighborhood he knew turn into a number, and the number he gives is about $45 billion in insured losses inside 48 hours. Before Palisades there was Lahaina. Before Lahaina, Paradise. Before Paradise, the Oakland Hills.
“There’s no reason why these fires should develop the way they are,” he told me. “And there’s no reason we should lose entire cities to an ignition.”
He decided that day to do something about it, and he started a company called FireBrake Interventions. He is its co-founder and its Chief Relationship Officer.
I did not invite him to YouXAI to talk about fire. I invited him because of what he built to keep the company alive. But the two turn out to be the same idea, and he says the same three words about both of them.
Every problem is easier when it’s small
Michael’s explanation of the fire technology starts with dishes.
“Everybody is happy to do one dish,” he said. “Well, most people are happy to do one dish. But facing the onslaught of a Thanksgiving dinner set of dishes is a daunting task.”
Fires work the same way. You cannot stop an ignition from happening. Somebody’s power line arcs, somebody’s mower hits a rock, and there is a spark. What you can do is deal with that spark before it becomes a firestorm, because once it is a firestorm the argument is over. “There’s only one way to stop a firestorm, and that’s to wait for it to burn out,” he said. “It’s a force unto itself, and it is absolute. It can’t be stopped. So the only way to stop it is to prevent it.”
The devices FireBrake is building are about a meter high. They sample the air around them constantly. Michael is careful with the word he uses next, and I liked him for the care. “I hate to use the word smell, because you need a nose to smell. But electronically, they can smell. And they can smell the difference between a barbecue and a burning tree.”
The devices wait for the recipe that produces a catastrophe. He walked me through what that recipe looked like on the day Palisades went up. Relative humidity the day before was well over 36, which is dry enough to be a red flag day and still wet enough to be survivable. By the next morning it was in the single digits. Add wind. Add one ignition. “Once you write that recipe, there’s only one thing you’re going to get, and that’s a firestorm.”
So the devices change the recipe. They raise the humidity and drop the temperature in a small region around themselves, and they do it the moment the air tells them to. There is nearly one for every house in a neighborhood, and each one talks to the other 700 or 1,000 around it.
Here is the part that matters, and it is the part that turns out to be about more than fire.
“While upstream you’re only putting out one ember, downstream you’re putting out thousands or tens of thousands of embers, which is thousands of fires. So there’s a huge effect upstream, and the farther upstream you can go, the better.”
Michael’s phrase for this is force multiplier. Act at the smallest possible scale, at the earliest possible moment, and one action does the work of ten thousand. He calls the company the first in the world to fight fires before they start.
The same idea, pointed at a spreadsheet
FireBrake is pre-revenue. It is three co-founders, one new intern, and a prototype they need about a million dollars to build. A million dollars is a rounding error to most venture funds, which is its own kind of problem, because a rounding error is not worth a partner’s afternoon.
So they are raising, and raising means cold outreach, and cold outreach has math attached to it. Michael put it plainly. A cold introduction gives you about a 3% chance of getting a conversation, “and that’s being generous. Probably more in a 1 1/2%.”
Work backwards from three or four real conversations and you land at roughly 230 firms. And you cannot simply pull a list of 230 names, because most of them are wrong for you and you will not know which ones until you have looked.
FireBrake sits at an intersection. It is fire tech, it touches insurance tech, and it touches environmental tech. So for every firm the questions stack up. Do they fund anything in those three areas? Do they write checks that small? Do they invest pre-revenue? Who have they already backed, for how much? Who are their own limited partners?
“You end up with 10, 15 data points per VC,” Michael said, “and each of those may have 10 or 15 data points as well. Exponentially, this is many, many months of work.”
He did the arithmetic on doing it himself. Forty hours a week, nothing else on his plate, a world he notes does not exist. Three months.
“And it literally did it in 14 minutes.”
The honest version of that number
Michael corrected the impression before I could form it, which is the reason I trust the rest of what he says.
Fourteen minutes did not produce 230 finished dossiers. Fourteen minutes produced the research. All the firms, all the partners, all the investments, all the connections, gathered and sorted against the criteria he wrote.
The dossiers themselves come out three at a time, and they come out three at a time on purpose.
“We do three a day, because that’s all we can handle. We’re a very small team.”
The throttle is the interesting design decision. He is not trying to send 230 emails. He is trying to have four conversations, and a conversation needs a human who has actually read the file.
The criteria he gave the system are specific in a way that most people’s prompts are not. This is who we are. This is who we’re looking for. And this is what a dossier has to contain. Every claim linked with a URL so his CEO can click it while reading. The email address on the page. The phone number on the page. He describes his own tone with it as a running reminder: “Oh, Mr. AI, don’t forget, you’re supposed to provide me all of that.”
Now he says one sentence and a sequence fires.
“I just go to it and I say, ‘I’m ready for the next three.’”
It writes the three dossiers. It sends them to Mike Ralston, the CEO. It loads them into the CRM. It creates a task for Mike to make the calls. Then it creates a second task, for Michael, to check that Mike made the calls.
“It’s just a whole cascade of things happens automagically.”
That last step is my favorite thing in the whole interview. He built the machine a follow-up on its own follow-up, because he knows how founders actually behave.
“I am not a thought leader”
I asked him the question I think most people would ask. Did he have Claude build him some sophisticated autonomous agent architecture?
He stopped me.
“I wanna make sure that I couch this correctly. I am not a thought leader. I’m a guy that has a startup that’s underfunded and has too few people working for it.”
Then he said the word he keeps coming back to.
“So out of desperation, I needed some significant force multipliers, and I needed them fast. And there’s no way for me to do it by myself. Generally, literally, physically, zero chance of that happening. So this is a desperate move on my part.”
He heard about Claude. He heard about Cowork. He decided to look before deciding.
There is a line he likes about that, and he could not remember who said it. “Contempt prior to investigation will hold a man in everlasting ignorance.” Then his own version, which is shorter and better: “It’s a good idea to know what you’re saying no to.”
What he found, he says, was a massive force multiplier. There is that phrase again, the one he used about the ember.
“I have a team of agents that answer to me now. I even have some agents that answer to another agent that answers to me. One has a managerial remit, one has a research remit, one has a proposal writing remit. And together they are my team.”
Three co-founders and one intern who started this week. Counting the AI his partners run too, he puts the working headcount at around 20.
Why the tool matters more than the trick
A big part of the workshop Michael is co-leading at YouXAI, with Mark from ModeForge, is about what AI does badly.
“AI is not a panacea for problems. AI is as imperfect as its creator, humans. And AI is as diverse as a tool in a toolbox.”
Then the line that will end up on a slide somewhere: “If somebody tries to eat a bowl of soup with a fork, they’re going to be disappointed.”
This is why he does not have one AI. He has several, and he chooses. He used Claude for the VC research because that is the job it fits. Another job might be Grok. Another might be Gemini. The skill is not knowing one tool deeply. The skill is knowing what you are holding.
And the model he uses for all of them is a person.
“Who has hired an employee where they did everything perfect the first day? Nobody. AI is the same way. It’s an employee. It needs to be trained, it needs to be worked with, coached. And it needs to be trusted less in the beginning than it can be at the end.”
That framing is doing more work than it appears to. It tells you the first output will be wrong and that this is not a reason to quit. It tells you the fix is coaching rather than a better prompt off the internet. And it tells you trust is something you extend gradually, based on what you have seen, which is exactly how the three-a-day throttle got designed.
What he actually wants you to take away
I told Michael I thought his workshop would be packed. I said something about the station attendants who push people onto the bullet train in Japan. He laughed, and then he got serious about the one thing he wanted people to hear.
“I’m not some software engineer. I haven’t been studying mathematics for the last 40 years. I’m a business guy. I’m a sales guy, primarily. And I was just desperate. I was looking for a solution, and somebody said this could help. I had no idea how much.”
Then the takeaway.
“My message would be, look, anybody can do this. Surely, if I can do this, if curiosity is all it takes to win this big with AI, then surely everybody has that modicum of curiosity. Now it’s just a matter of actually doing it.”
The hair on my arms stood up while he was telling me all of this, and it was not the fire technology that did it. It was watching the same idea run both halves of one person’s business without him ever pointing it out.
Go upstream. Act while the thing is still small. One ember, or one afternoon writing down exactly what you want, and the thousands of fires downstream never happen.
Michael is at Startup Waco most Wednesday mornings at nine, just down Austin Avenue, arguing about how to make Waco better as a business town. He came here from California to build something. That is a good problem for a city to have.
He and Mark run their workshop at YouXAI on September 12.
Tickets are $79 through August 31, then $99, at youxai.live.
— Fernando Labastida YouXAI Speaker Preview Series