An agent that keeps thinking after you stop typing — and the whole harness is under 10,000 lines of Bash
The Laude Institute, working with MIT, has open-sourced Headlong, an agent harness built on a different premise from the usual one. Most harnesses are reactive: you send a task, the agent finishes it, and it sits frozen until you come back. Headlong runs a continuous self-guided loop modelled on inner monologue — a message from a person does not start a session, it is one more observation dropped into the thought stream. The agent has a name and a personality, sets its own priorities, starts its own projects and pings you when it has something to say. The core is under 10,000 lines of Bash under Apache-2.0, and the published demo shows an agent finding and fixing a bug in its own recall code over 48 minutes with nobody watching.
Why it mattersPersistent agency is the part of the agent story everyone assumes is coming and almost nobody ships, because an agent that never stops thinking burns tokens and can wander somewhere useless. A small, readable, openly licensed implementation is the cheapest way for everyone else to find out what actually goes wrong before betting a product on it.
Read in the app — free, in 9 languages
Related stories
Nvidia says its new rack does 30 times more agent work per megawatt — it ran the test itself, and the benchmark's authors have not signed off2026-08-25The new Mac Studio holds 512GB of memory the GPU can address directly — a rack's worth of model, under a desk
2026-08-25ByteDance has folded its coding tool and its agent builder into one app, and pointed it at the office
2026-08-25Rumour: developers say the anonymous model beating GPT-5.6 on OpenRouter is Zhipu's — the evidence is a tokenizer and a stack trace
2026-08-25Your 'local' AI image asks a Microsoft server for permission first — and comes back stamped with an ID only that server can read
2026-08-25