Tyler Edwards

Tyler Edwards co-founded Overmind in London in July 2025 and is its chief executive. Before that he spent eight years building AI systems for British intelligence agencies. His co-founder Sam Brunt is chief operating officer.

A £2m seed round

Overmind came out of stealth on 11 February 2026 with a £2m seed round led by Osney Capital, a specialist cybersecurity investor, with 14Peaks, Portfolio Ventures, Antler and Endurance Ventures. At launch it described itself as a supervision layer for AI agents: software that watches how agents behave in live systems, detects deviations and lets teams step in, using reinforcement learning to improve the agents over time. It set out to sell first to regulated sectors such as legal, healthcare and fintech.

“The AI security industry is trying to secure the wrong thing. Model security matters, but the deeper challenge is operational,” Edwards said in the announcement.

From supervision to training

By September 2026 Overmind was presenting itself as a model training platform for AI teams. It connects to a company’s codebase and to observability tools such as Langfuse, Braintrust and Datadog, turns the production traces of its agents into cleaned training datasets, generates evaluations as the code changes and trains specialized models that the customer owns. Overmind says small models trained this way can beat frontier models on accuracy, hallucination and cost for specialized tasks, an argument Edwards makes in his post “Think Smaller”.

The two positions share a premise: an agent’s behavior in production is the best evidence of what it does, and that evidence can be used to make the agent better as well as to police it.

Why it matters to UK buyers

For UK businesses running AI agents on frontier models, Overmind’s proposition is that repetitive agent tasks can move to a smaller model the business controls, with lower running costs and more say over where its data goes. That is most relevant in the regulated sectors Overmind first targeted, where control over how an agent behaves is a compliance question as much as a technical one.

Edwards has moved quickly, from incorporation to a funded launch in seven months and to a sharper product within a year. The change of direction is a bet that owning the model, not just watching it, is where control of AI agents will end up.