AutoEnvScaling: agents build their own training tasks for human-free self-play RL
_AndrewZhao · x · 2026-10-14
Recursive self-improvement needs agents that can build their own training environments, but researchers remain in the loop, limiting scale.
The paper introduces AutoEnvScaling, which turns the Data Flywheel into a terminal task: a proposer agent constructs complete Harbor tasks and sandboxes, working much like a human researcher. This enables self-play RL for terminal agents — one model improves as both builder and solver, with no human in the loop.
Related event: AutoEnvScaling Lets Agents Build Their Own Training Environments(3 posts)→
More from coding & agent
- Omarchy bug: missing rtkit causes Chrome/Electron apps to get SIGKILLed — fix inside — mark_k · 2026-10-15
- 2,611 real interviews show rater bias predicts AI agent scores better than actual performance — Exp_Mark · 2026-10-15
- Google's DuoDiscover tops NVIDIA SOL-ExecBench, 2.68x faster than production TPU kernels — lateinteraction · 2026-10-15
- Agent stack check: harnesses, OTel telemetry and MCP are settled — evals remain a mess — _ScottCondron · 2026-10-15
- Claude Code's safety guardrails refuse even non-sensitive env var copies, user complains — Promptmethus · 2026-10-15
- Fragment raises $12M Series A to make AI agents safe financial actors — mattturck · 2026-10-15