notes on 'How to do data' from th Zai team that others should steal as they build their own environments - 'Research agents collect task pat…

notes on "How to do data" from th Zai team that others should steal as they build their own environments - "Research agents collect task patterns from real work and turn them into runnable long-horizon environments" good environments come from faithfully simulating the real world. You already own a lot of that data in your Traces. Agents can use those + other data to bootstrap environments everyone can own this process for themselves by carefully using coding agents - "these pipelines still require a meaningful amount of human-in-the-loop" basically...look at the data one of the highest value things builders can do is selectively infuse their domain knowledge into the environment building process, this means inspecting rollouts and verifier structure to make sure it aligns with stuff they care about Agents help automate a lot but without humans the synthetic data pipeline doesn't worth end to end....yet Introducing GLM-5.3: Built to Code. Ready for Cyber Defense. - Top-tier coding and agentic capabilities, achieved through post-training on the 743B base model - A major leap in cybersecurity, setting a new standard among open models Tech Blog: https://z.ai/blog/glm-5.3

Related

• Evals ~= Environments…they’re one of the best investments a team can make for improving agents Step 0: Turn On Tracing for Agents Step 1: Po…
• auto-research style proposal loops should be data driven! they largely work best only when Data/Evals/Feedback give a useful gradient to hil…
• build v1 of agent ship it (dogfooding counts) ⭐️ collect tracing data ⭐️ ⭐️⭐️ point agentic compute at data ⭐️⭐️ understand failures at scal…

Source: Harrison Chase (X) | 2026-08-14