Jekyll2026-08-19T05:06:00+00:00https://heye.me/feed.xmlHe YeHe Ye — Lecturer (Assistant Professor) at UCL, building the next generation of code agents for software engineering.He Yehe.ye@ucl.ac.ukThe Meta-Agent Challenge: Are Current Agents Capable of Autonomous Agent Development?2026-08-12T00:00:00+00:002026-08-12T00:00:00+00:00https://heye.me/posts/2026/08/meta-agent-challengeTakeaway
A code agent gets a sandbox, a hidden test split, and 12–24 hours to write the agent that solves the task; across five domains, meta-agents rarely beat human-engineered scaffolds and the design process is brittle.
Remaining work
Grading is rule-based. Under an LLM judge, how fast does the meta-agent optimize the judge instead of the answer?
The executor is weak (Qwen3-8B). What changes with a strong one — or when designer and executor are the same model?
This and Meta-Harness are both single-domain RSI. What does cross-domain RSI look like?
]]>He Yehe.ye@ucl.ac.ukMeta-Harness: End-to-End Optimization of Model Harnesses2026-08-09T00:00:00+00:002026-08-09T00:00:00+00:00https://heye.me/posts/2026/08/meta-harnessTakeaway
Full trajectories, not summaries
A coding agent does the evolving
It’s a search problem; the hard part is credit assignment
Code-as-representation is a free regularizer
Remaining work
Credit assignment stays qualitative (no per-component score), and no guarantee each round improves.
Large, open-ended search space explored under a tight budget — running each candidate on the frozen model is the compute bottleneck.
Generalizes across datasets and models, but not across task types.