Harness adapters: per-agent-framework LoRAs cut tool calls, $5 for 1,350 eval sessions
LysandreJik · x · 2026-11-06
Liquid AI's Lysandre shares progress on making models work better with agent harnesses: "harness adapters" — small LoRAs trained with RL inside the actual harness they'll be used with (pi, opencode, hermes, etc.).
The insight: the same base model behaves very differently depending on surrounding tools, prompts, and control flow, so specializing a tiny part of the weights per environment beats one-size-fits-all weights.
Early results are encouraging — adapters make fewer tool calls than the base model while keeping or improving accuracy on held-out tasks. And since they're LoRAs, operations are cheap: one LFM2.5-24B-A2B experiment evaluated 9 checkpoints / 1,350 agent sessions on a single H200 in under an hour, for about $5 total.
More from coding & agent
- Agent builds cable modem emulator that speaks DOCSIS in Rust and boots unmodified firmware — moyix · 2026-11-06
- ChatGPT's 1.2B weekly users are already doing agentic tasks in main chat — omooretweets · 2026-11-06
- bas-guard: a local offline security scanner giving Claude Code setups an "antivirus for agents" — victor_explore · 2026-11-06
- TMail Open-Sourced: Private Email for AI Agents on Your Tailscale Network — evilsocket · 2026-11-06
- wesbos migrates 15-year WordPress site to Emdash with agent skills, page speed jumps 52 to 75 — ritakozlov · 2026-11-06
- LLVM Dev Meeting 2026: Building durable open-source infrastructure in the agentic era — clattner_llvm · 2026-11-06