Founder
Mike Smith
Nothing is true until something tried to break it and couldnt. Small Mind is where that finally pointed at AI.
My grandfather was a tool-and-die maker and a cabinet maker, and he is where all of this actually starts. When he built something it was square, it was exact, and it was built to outlast him. He taught me the thing the whole company is built on before I had any words for it: don't trust the smooth answer. A joint that looks finished isn't finished. A cut that reads clean can still be wrong. You check it, because looking right and being right are two different things, and the gap between them is where everything fails.
I was a kid who took everything apart to see why it worked, then tried to build one that worked better or make it do things it wasn't meant to. That never stopped. Over the years it turned into a habit of picking something hard and learning it hands-on until I understood it from the inside: engines, to the point full drift builds with no mechanical background. Guitar, and clawhammer banjo before it. Carpentry and cabinetmaking, my grandfather's trades. C, then Python, then the whole AI stack around them. A complete 2D platform RPG, built engine up. Different subjects, same move every time, and every one of them taught the same lesson again: the thing that looks done and the thing that is done are not the same thing.
I have also been on the internet since I was ten, which turned out to matter more than it should. I watched the whole thing unfold in real time, platform after platform, and watched what happened to the truth as it did: one source becoming forty copies, confidence detaching from evidence, the loudest version winning and the sourced version getting buried. I learned to read a thread for the one comment that actually knew something, and to distrust a claim precisely because everyone was repeating it. That was provenance instinct too. I just didn't have the word for it yet.
Then I spent my career in rooms where being wrong costs money that afternoon. I led a field sales organization of about 550 people to record numbers and pioneered the channel it ran on. I managed a plant floor for a just-in-time production company, which is the environment that hammered the lesson home hardest: in JIT, a number that can't show where it came from stops a line, and a smooth wrong signal cascades through everything downstream before anyone catches it. Then a role at one of the big automakers. Different floors, same discipline every time. A gauge that reads fine but isn't wired to anything will pass every glance and fail you exactly when it counts.
So when I turned all of that on AI, the thing that bothered me was obvious and everywhere: confidence with no receipts. So now the thing Im breaking is AI. I red-team agentic systems, the ones you hand your data to and let loose to act for you, and I go looking for the spot where some attacker controlled text, a poisoned email, a document with something buried in it, turns the agent against the person who owns it. Indirect prompt injection. Its the hole in every AI agent shipping right now and most of them have no real answer for it, they just take it on faith that the model behaves. I go after the scorekeepers too. I took the JailbreakBench leaderboard, kept every attack exactly the same, and changed one thing, the model doing the grading. Four of five graders reshuffled the whole board. So the ranking everyone points at depends on which judge you picked, not on what actually happened. That study is public and preregistered. Looking right and being right are two different things and right now most of AI security is just looking right.
So I built a company around that space. Small Mind builds AI that has to earn its confidence, and breaks the AI that hasnt. Two arms, one rule. Nothing is true until something tried to break it and couldnt.That is Small Mind.
Reach me
Small Mind LLC, Michigan. Work and receipts at github.com/Threadborne. Email smallmindllc@gmail.com.