Teknium runs Hermes Agent for 15 hours with recursive subagents, shedding 375,000 lines of code
Teknium · x · 2026-09-03
Teknium stress-tested Hermes Agent by giving it a single /goal: clean up its own repo. It ran for 15 hours and simplified, unified, optimized, or removed enough code to shed 375,000 lines.
- Waves of 120 subagents (3x sets of 15) ran in parallel for nearly the whole duration, with subagents spawning their own subagents recursively.
- The entire workload ran on a single desktop machine, which he calls an epic test of the limits of the latest model generation and the agent harness. More details promised soon.
Related event: Hermes Agent runs 15 hours, deletes 375K lines of code(4 posts)→
More from coding & agent
- LukeW built an agent-orchestration app for a foldable: managing agents is now an all-day job — LukeW · 2026-10-21
- Agentic SDLC: engineers shift from implementers to agent orchestrators as cycles shrink from weeks to hours — Pavan_Belagatti · 2026-10-21
- OpenAI's dots can now use company tools via Merge for Workforce with employee-bound permissions — shensi · 2026-10-21
- The topology of an LLM agent: where the model, loop, and tools run shapes cost and security — davidcrawshaw · 2026-10-21
- Agents with full-disk and accessibility access are an enterprise security nightmare, argues Runlayer — leveragedupside · 2026-10-21
- Acellera runs an unsupervised virtual lab of AI scientists on a real IL-23 molecular discovery program — gdefabritiis · 2026-10-21