It feels like agent harness evolution runs on two axes that usually get conflated. There’s the temporal axis: simplify as models improve, st…

It feels like agent harness evolution runs on two axes that usually get conflated. There’s the temporal axis: simplify as models improve, stripping components that compensated for limitations the new model no longer has. Sonnet 4.5 needed context resets; opus 4.5 didn’t. Opus 4.5 needed sprint decomposition for planning; opus 4.6 didn’t. Then there’s the lateral axis: tune the harness per model family, because different models need different tool names, prompt framing, middleware, and orchestration at any given point in time. Model labs mostly see the temporal axis because their harness only talks to one model family. But if you’re building in a multi-model world with models from anthropic today, openai tomorrow, google next week, and an OS lab after that, you probably should care about both. Harness profiles are a mechanism for that second axis, they allow you to keep the agent abstraction stable, but let the model-specific harness contents move. https://docs.langchain.com/oss/python/deepagents/profiles

Source: Harrison Chase (X) | 2026-05-06