AI Research Engineer (Staff)

An opportunity to build the environments that let AI agents learn from doing, not just from data. A young, extremely well-backed AI company whose work feeds directly into the top research labs. Real technical ownership at the most senior IC levels - with your hands still on the work. The frontier is shifting from training on fixed data to learning through experience. Our client builds the software worlds where AI agents practise real tasks - from simple web apps through to full enterprise systems - stumble, and get sharper with every attempt, producing the hands-on training signal that enhances models further. That work goes straight into the research pipelines of the top labs. They're now adding senior Research Engineers who can set direction across the hardest AI problems while staying deep in the work - people who lead through judgement, not job title.

What You'll Be Doing

• Take a major AI or research-engineering area and carry it the whole way - from framing the problem and choosing the direction to something that moves in production.
• Break vague, company-sized questions down into concrete experiments, priorities, and a clear way to tell whether they're working.
• Drive how training data gets made and curated, how models are fine-tuned and post-trained, and how agents are measured and improved.
• Build and refine the harnesses, sandboxed execution and feedback loops that let agents run, get scored, and improve inside interactive workflows.
• Decide where the leverage is: which experiments to run, what 'good' looks like, and when to double down, scale, or walk away from an approach.
• Get promising prototypes out of research and into dependable, well-instrumented production, working shoulder to shoulder with the software engineers.
• Steer big, multi-person efforts through the strength of your calls rather than a reporting line - lifting the quality of everyone else's technical decisions while keeping your own hands in the code.

What You'll Need

Essential

• You've trained, fine-tuned or post-trained models yourself, and you know how to tell whether what you did actually moved the needle.
• A real command of modern ML - post-training, reinforcement learning, optimisation and evaluation - not just passing familiarity.
• Genuine software-engineering depth: you write, review and defend production code, not just prototypes.
• A history of taking messy, ill-defined problems and turning them into something that ships.
• You've run significant technical efforts over months, carrying several people with you without formal authority over them.
• Low ego, and a deliberate choice to stay an individual contributor rather than manage.

Bonus

• You've built serious agents, agent harnesses, evaluation stacks or sandboxed execution infrastructure.
• You've owned training-data or synthetic-data pipelines at real scale.
• Something you trained or fine-tuned set the bar in a domain that mattered.
• You've managed people before and chose to come back to a hands-on senior track.
• A strong research or academic grounding - a plus.

The Person

You want the authority to shape hard technical direction without being pulled away from building it. You're at home when the goal is clear but the route isn't, happy to commit on partial information and change your mind quickly when the data says so. You move easily between reasoning from first principles and just shipping the pragmatic thing, and you'd rather make the engineers around you sharper than outrank them.

The Role

Around two days a week in the office. Hired at Staff or Principal, with the level set by the ownership and judgement you show through the process. Strongly competitive salary and meaningful equity. Hands-on at every level - you stay in the code, the experiments and the calls.