Agentic teams
We build teams of AI agents that split a workflow between them, check each other's work and hand off to people when it matters.
The problem we solve
A single chatbot can answer questions. Real work, like onboarding a supplier or resolving a claim, takes several steps, several systems and a few judgement calls.
We design teams of agents the way you would design a team of people: each with one job, the tools to do it and a clear hand-off. A person approves what matters, and every run is traced so you can see who did what.
Typical engagement
- Workflow design (1–2 weeks): the roles, the tools and the points where a person decides.
- Build (6–10 weeks): the agents, their evaluations and the dashboard your team uses to supervise them.
- Run: monitoring, cost control and new roles added as the team proves itself.
What we deliver
- A workflow broken into roles an agent can own
- Agents with the tools, data and limits each role needs
- Hand-offs between agents and to the people who approve
- Evaluation sets that test every agent before release
- Traces of every run, so you can see what each agent did
Tools we use
We pick tools to fit your systems and your team, and these are the ones we reach for most.
Python
TypeScript
Anthropic
OpenAI
- LangGraph
- Vector databases
Related work
- Automated testing platform
An internal app that automates testing workflows, stores results and uses machine learning to predict where tests will fail.
- Property auction & asset management
An auction and asset management platform with live auction workflows, asset listings and reporting, fed by ETL data pipelines.
- Counselling platform MVP
A therapy-adjacent platform with affirmation tracking, interaction logging and a recommendation engine that personalises each user's journey.