Avoid humans where they add nothing. Keep them where they decide.
Handing engineering work to agents needs more than a prompt. Agents need limits, a map of the systems they touch, and checks against what those systems recorded. Risky changes should wait for a person. We started Just Avoid Humans to build that.
How we build
- Accountability
Agents do the work. People answer for it.
An agent can be responsible for a task, never accountable for it. People set the purpose, the limits and when agents get more room. They review the source, not just the agent's summary.
- Proof
No promises. Proof you can check.
We assume a model can be manipulated. So a prompt is not a permission: we design our controls as code that sits around the model.
- Exit
Designed so you can leave.
Our products are designed to run in your cloud account, with keys you hold, and to keep their records there too. We test the exit before anything depends on it.
What we're building.
One idea, three products: a source on every action, fact and answer.
- WorkforceAgent runtime for engineering workIn build
Runs agents on engineering tasks, each with scope, budget and rules for when a person must step in.
- CensusContext layer for agentsIn build
Maps your company from your sources and gives each agent the part its task needs.
- EngineGoverned answer engineClosed pilot
Answers questions about your data in plain language, with a source on every sentence.
People stay in charge. Fewer steps need someone watching.
Let's talk about your experience with coding agents.
How your team uses them today, which steps you still keep for a person, and what happened the last time one went wrong.
Book a 30-minute call