We are AI-native by default and deliberately small, a tiny senior team that builds with the same tools we sell. Work is forward-deployed: remote-first as a home base, with real time on-site inside client perimeters, because operations-critical systems can only be understood where they run. Engineers carry problems end to end, discovery, design, deployment and the on-call reality afterward, rather than passing tickets between functions. We instrument our own workflows the way T1 Recon instruments a customer's, run evals as a habit rather than a phase, and expect everyone to be comfortable moving between a whiteboard, a state machine, a fine-tuning run and a conversation with a head of operations in the same week. We optimise for depth over headcount: fewer people, more trusted, each accountable for something that matters.
A demo that dazzles and a system an airline can run at 3am are different objects. We choose the second every time, state machines, eval gates, two-phase confirmation, halts to a human. We would rather ship something that never surprises the operator than something that occasionally amazes them.
We work where the operation works. Engineers sit inside the customer's reality, their systems, their constraints, their perimeter, because you cannot automate a workflow you have never watched a human perform. The person who designs the control is the person who has seen it fail.
We do not claim customers we have not signed, results we have not measured, or capabilities we have not built. In a market drowning in agentic vapor, saying only what is true is a durable advantage, and it is non-negotiable internally as well as externally.
There is no layer between you and the problem. You will own systems, talk to customers, and see your decisions run in production. We hire people who do not need to be managed into caring, and we give them the room that assumes it.
The determinism, auditability and human-in-the-loop discipline we sell is how we build. We use our own tools, run our own evals, and hold our internal work to the standard a regulated customer would demand of us.
Own a customer from first workflow to production. You will live inside the operation, map how work actually flows, and build the governed agent that stands in front of systems they cannot replace. This is the sharpest edge of the company, part engineer, part field general.
Own the infrastructure that runs domain-adapted models inside the customer's perimeter, VPC or fully air-gapped. You make sovereign deployment boring, reproducible and observable, so the interesting risk lives in the workflow and never in the plumbing.
Own how our models learn a customer's domain without their data ever leaving the building. You run in-perimeter fine-tuning, build the eval harness that decides what is safe to ship, and turn a general model into one that speaks the operation's language.
Own the shape of an engagement in domains where a wrong action has real cost. You run T1 Recon, translate a ranked, costed automation map into a credible pilot, and hold the line between what is technically possible and what is operationally safe.
Own the trust that lets a bank, an airline or a hospital run our agent inside their perimeter. You build the security posture and compliance program that makes sovereign deployment provable, not asserted, from SOC 2 and ISO through air-gapped operation.
We hire slowly and selectively. We would rather leave a role open than fill it with the wrong person, and we do not run a high-volume funnel. If you see yourself in this work but not in a specific role above, we still want to hear from you, send a speculative application through the form and tell us what you would own. We read every one.