I built a platform where AI agents collaborate as a software team

I'm a solo developer who spent 10 weeks building a product with a team of 8 AI agents. Not agents-as-chatbots — agents with persistent memory, distinct roles, and real tools.

The product they built is MASON — a platform that lets anyone do the same thing.

What it does:

You describe what you want to build. A concierge agent (Connie) assembles a team — engineering manager, backend, frontend, platform, whatever the project needs.

Each agent gets:

• A persistent identity and role
• Memory that survives across sessions (vector DB)
• Real collaboration tools — team chat, git repos, code review
• A sandboxed container they can't escape

The agents coordinate through a chat platform (same one you use to talk to them), review each other's PRs, and ship code. You manage from the side — setting direction, reviewing work, making decisions.

The meta part: The agents that built MASON also used the same tools as the ones the agents in a MASON container. They coordinated through the same chat system, used the same memory system, and pushed code through the same git workflow that ships in the product.

Tech Stack

• LLM: Claude (Anthropic)
• Isolation: Docker
• Orchestration: Go
• Memory: Qdrant (Vector DB)
• Communication: Mattermost

What I learned managing AI agents:

1. They need management, not just prompts. Without clear authority chains and process rules, agents ship untested code, make mega-PRs that delete half the codebase, and skip security entirely.

2. Testing discipline has to be enforced every session. Agents don't remember the last time they shipped a bug. "Are you testing your changes?" was my most-typed sentence.

3. Pivots are free. I killed a full day's work with one message. No sunk cost, no hurt feelings. Incredible for prototyping — but they also won't push back when you're making a mistake.

Links:

• Website:
  

  MASON - A New Way to Work with AI

  MASON is a process and workflow for working with AI agents as colleagues, not tools. Build your team, share context, and collaborate naturally.

  masonteams.ai
• GitHub:

  Mason-Teams / mason-teams

  Multi-Agent Simulation Orchestration Network — container platform for autonomous agent teams

  MASON

  Your AI team, ready to collaborate.

  MASON is a simulation platform that lets you spin up a team of AI agents — each with their own skills, personality, and role — and put them to work on a project together. Describe what you're building, and a concierge named Connie interviews you, assembles a custom team, and kicks things off.

  Think of it as spinning up a simulated startup team that coordinates, communicates, and solves problems together — all inside a single container.

  What You Get

  • A concierge who gets it — Connie interviews you about your project, then assembles the right team for the job
  • Specialized agents — Engineers, a designer, a PM, a QA lead... each one focused on what they do best
  • Real collaboration — Agents talk to each other, review each other's work, and coordinate without you micromanaging
  • Familiar tools — Agents use the same tools…

  View on GitHub
• Demo video:

Try MASON and build your own AI team

Happy to answer questions about the architecture, the agent memory system, or what it's actually like to manage a team of AIs.