The shape of it
Most automation asks you to build the workflow: draw the boxes, wire the arrows, maintain it forever. Honey Badgers inverts that. You describe the outcome you want and the system works out the steps, then keeps adapting them as it learns what you accept and what you send back.
Everything lives inside a project. A project has its own memory, its own rules, its own credentials and its own team of agents. You can run many projects in one workspace, and they stay isolated where it matters.
The Project Leader
Every project gets one Project Leader agent. It is generated from your description of the work and it knows two things: what you are trying to achieve, and which tools it actually has available. That second part matters — it plans against real capabilities rather than inventing steps it cannot perform.
The Leader does not do the work itself. It decides what needs doing, in what order, and by whom.
The team it hires
The Leader assembles specialists as the work demands them — a researcher, a writer, an editor, a publisher. Each one is a separate agent with its own instructions and its own engine, so a cheap fast model can grind through the routine parts while something stronger handles what goes in front of a customer.
- Specialists are per project. A tone that suits one brand does not leak into another.
- Teams change over time. If the work shifts, the Leader can bring in a role that did not exist when you started.
- Every agent is inspectable. You can read what an agent was told and what it did, rather than guessing at a black box.
Memory and rules
Two things stop agents from repeating mistakes. Memory is what the project has learned — what you approved, what you rejected, what worked. Rules are what you have told it explicitly: never say this, always include that, publish nothing on a Sunday.
Rules are editable and reviewable rather than buried in a prompt, which means when behaviour is wrong you can go and change the thing that caused it instead of re-explaining yourself every run.
This is the part that makes the difference between a chatbot and something you can leave alone. A chatbot forgets. A project accumulates.
Plans and approvals
Work surfaces in two feeds. The Plan feed shows what the team intends to do before it does it. The Publisher feed shows what is finished and waiting to go out.
You decide how much rope to give it
Approve item by item while you are still building trust, then loosen it once a project has a track record. Nothing publishes behind your back unless you have said it can.
Four ways to reach it
- The dashboard — the full picture: feeds, agents, memory, rules, schedules.
- Chat — talk to a project the way you would brief a colleague.
- The CLI — drive it from a terminal, script it into whatever you already run.
- MCP — point your own AI tooling at it and let that drive Honey Badgers directly.
Where the AI comes from
You bring it. Honey Badgers is the orchestration layer — attach a Codex subscription you already pay for, a Claude API key, or both, and set which engine each project and agent uses. We do not mark up your tokens.
The full economics, including what each subscription tier costs and where the floor and ceiling sit, are in the bring-your-own-AI guide.
Want to see it on your own work?
Tell us what you would put a team on and we will tell you honestly how it would be set up — and whether it is worth automating at all.