AI agents that hand back work you can grade
Hand recurring work to agents. Keep the ones that get better.
Tell a foreman what you need. It starts specialist agents, they hand back files you can open and compare, and you grade them. Every save of an agent is a version, so the grades show which changes helped.
Sign in to startRuns on your own model keys: Anthropic, OpenAI, Google, DeepSeek or OpenRouter.
1
Ask a foreman
Describe the job in a chat. The foreman asks what it needs to know, then decides which agents do what.
2
Agents do the work
Researchers, writers, designers, coders: each with its own model and tools, and its own computer when the job needs one.
3
Grade what comes back
Open what they hand over, compare drafts side by side, and grade them. Grades add up on the agent version that made the work.