Agents That Learn as They Work

Two releases point the same way: agents that get better without a new model.

Google Research has open-sourced Harness², @kimmonismus reports. The framework lets AI agents improve their own prompts, tools, memory and workflows. It tests different configurations and carries useful lessons into future tasks, while the underlying model stays unchanged.

@kimmonismus notes that “an increasing number of releases concerning self-improving models are being published.”

@sudoingX shipped xpress agent v0.1.0. It is local first, “learns as it works” and is written in Rust. A day earlier it was a post asking a Grok Bot to build it. The Grok Bot prototyped it, and @sudoingX then took it into Cursor and “engineered every inch of it.”

Since last night it is the only agent he opens, he says, and it “turned out better than i expected.” He is asking followers whether they would use it if he open-sources it.