news.inema.pro

news.inema.pro / All news

agents Published on

EXECUÇÃO LONGA — MODO CONTÍNUO

A discussion of AI agents that work for hours or days, with a `/goal` prompt protocol for ongoing tasks and context on autonomous agents.

Long-running execution is a way to organize AI agents to work on tasks that may take hours or days. Instead of responding to just one request, the agent follows a defined goal and a cycle of analyzing, executing, testing, correcting, updating, and continuing. OpenAI Dots is presented as an example of a permanent digital assistant with its own computer, browser, and tools.

For users, the change is the ability to follow longer projects without starting over in every session. The agent records the goal, plan, state, and progress in external files, which help it resume the task after the context is compacted. It can also make reversible, low-risk technical decisions; real blockers, destructive risks, and business decisions are left for human review.

To get started, define a goal that can be verified and a clear condition for stopping. In Codex, the `/goal` protocol brings these instructions together; permanent instructions in `AGENTS.md` avoid repeating the process in every session. Check for progress, test, and issue records, and make sure the agent asks a person when it encounters a blocker or a decision that requires human judgment.

Read at the source →

The link opens the original content; some require a subscription.

Go further on INEMA.CLUB

More news

  1. 13/11
    refazvideo: turn a good English video into an explainer in PortugueseIntroduces a method using an AI agent to remake a video in another language in Portuguese with the user's avatar, without needing to record anything.
  2. 13/11
    Ficha do Enxame: organize AI agents in Google's /swarmIntroduces the Ficha do Enxame kit for Google Antigravity's `/swarm`, with videos in basic, technical, and specification versions.
  3. 11/11
    Audit your Claude: find old instructions that get in the way in Fable 5.2The project helps find old instructions that may get in the way of Claude in Fable 5.2.
  4. 10/11
    Painel de Agentes: organize your agents on one screenThe free kit brings agents, Claude Code sessions, bots, timers, and queues together on one screen so you can organize and talk to them on your computer.
  5. 10/11
    Caixa do Agente: your AI agent in a boxThe free kit runs Claude Code, Codex, and DeepSeek Harness in a disposable Docker box.