DeviDevs, internal system

Our company runs on AI agents that split the work and check each other

DeviDevs runs daily on a multi-agent system built on Claude. A manager agent takes the request and splits it between specialised agents: one writes code, one does research, one writes copy. An independent reviewer reads their work before anything leaves the company. We build the same pattern on your company's processes.

Updated

The challenge

One agent asked to write code, do research and write copy makes mistakes without noticing, and nothing stops it.

What we built

We split the work into roles: a manager that plans, agents that write code, research or write copy, and a reviewer started separately from the author whose job is to find what is wrong. Before deciding, the agents check the company's memory.

How it works

  1. The request reaches the manager agent, which breaks it into tasks.
  2. Tasks go to the right agent: code, research or copy.
  3. The reviewer reads the result and looks for what is wrong.
  4. Anything that touches money, contracts or clients waits for a person's approval.

What changed

  • Our inbox and code go through the system daily, and invoices when we issue them.
  • One agent's work is read by another before it leaves the company.
  • A person gives the yes where a mistake would cost something.

Stack

  • Claude
  • Claude Code
  • MCP
  • Python

Related services

Related projects

Want agents working like this on your company's processes?

Tell us what your team does by hand today. We will come back with what agents can take over and what stays with a person.

Talk to us about an agent for your company