Skip to content

Research

AI Agents

Tool use, multi-step planning and multi-agent orchestration; MCP integration, human-in-the-loop workflows and safety guardrails.

AI Agents

What we work on

  1. 01Tool use and function calling, multi-step planning and task decomposition
  2. 02Coordination and division of labour across multiple agents
  3. 03Connecting to existing enterprise systems through MCP-based integration
  4. 04Human-in-the-loop workflows: which step requires approval, permission boundaries, reversibility
  5. 05Agent evaluation: how we measure success on multi-step tasks and how we catch regressions

Why we work in this area

There is a large difference between a language model producing text and a system acting on your behalf. The second, when it gets something wrong, may produce consequences that cannot be undone. In agent systems the real engineering problem is not the model's intelligence but where authority stops and how errors are caught.

The question organisations ask most in this area and find answered least is this: what happens if this system gets it wrong? The answer to that question sits at the centre of our work.