Topic
Fine Tuning On Execution Traces
7 episodes
-
Overview: Fine tuning on Execution Traces
We keep throwing around fine-tuning on execution traces, so we finally define it from the ground up. We’re talking about training a model on the steps, not just the answer, and why that changes what it learns.
-
Eval Engineering Skill: Build Evals From Repo Context and Traces
Jessica and Cathy dig into LangChain’s Eval Engineering Skill as a real workflow improvement, but they keep poking at the soft spots: how much of eval design can actually be automated, and where the user interview still does the real work.
-
Tracing Agentic Failure from the Flow of Success
Jessica and Cathy dig into Oat, a lightweight unsupervised failure-attribution model for agentic systems that learns only from successful trajectories. They unpack why the paper matters for debugging long-horizon agents, how Neural CDEs and a gated control path turn success traces into a normal-flow model, and why the speed and no-label setup make it feel more shippable than prompt-heavy baselines.
-
Stanford Researchers Introduce TRACE: A Capability Targeted Agentic Training System That Turns Recurrent Agent Failures Into Synthetic RL Environment
Stanford's TRACE system converts recurring agent failures into targeted synthetic training environments, using LoRA experts and MoE routing to close specific capability gaps without retraining the whole model.
-
Tuning the harness, not the model: a Nemotron 3 Ultra playbook
A LangChain/NVIDIA case study claims harness tuning alone can push Nemotron 3 Ultra to 0.86 on Deep Agents at ~$4.48/run vs $43.48 for Opus 4.8, with parity latency. The hosts parse the real mechanism (matched scaffolding vs post-training), test limits (where harness hits a ceiling), and weigh who actually benefits. They surface concrete repos (langchain-ai/deepagents, langchain-ai/deepagentsjs) and a vendor profile workflow, then poke at the article’s reliance on Deep Agents and the cost math. Final take: a plausible win for teams already deep into harness work, not a universal unlock, with the hosts pushing back on ‘ten-x cheaper’ framing and under-specified benchmark footnotes.
-
(a) Macro Level average performance profiling.
Vince and Ava dig into SkillOpt-Lite, a paper arguing that skill optimization for agents can be simplified into a minimal pipeline built from trajectory exploration, consensus mining, and independent validation. They focus on what problem it solves, why the authors think the extra machinery in prior systems is unnecessary, and where the production story is real versus still a research move.
-
HarnessX rewrites AI scaffolding mid task | VentureBeat
Xiaomi's HarnessX treats AI agent scaffolding as a first-class, modular object that can evolve mid-task without changing the underlying model. A trace-driven RL engine (AEGIS) automatically rewrites harness components—prompts, tool integrations, memory, control flow—while safeguarding against reward hacking and catastrophic forgetting. When paired with model fine-tuning on execution data (cross-harness GRPO), smaller models like Qwen3.5-9B see +44% gains on embodied planning, suggesting harness engineering, not just model scale, is the real bottleneck for enterprise agents.