Set up an always-on company task agent

They built a self-running agent that acts like a company employee and uses long context, a diary, and markdown files to stay oriented.

Done with CodexChatGPTยท Advancedยท Apr 2026

What they did
The developer chose to build their own autonomous setup instead of using OpenClaw. It runs with Codex on a ChatGPT Plus account, starts through systemd, and uses plain UNIX email plus a regular user account to take on work. To keep track of things, it relies on one rolling context, a diary, and a directory of markdown wiki pages.
How it went
They say it has handled many kinds of company tasks and has not shown memory issues yet. They also note it is still early, and if the workload becomes too much, they would split duties across extra bots rather than stretch one agent further.
Tips
A setup like this needs a clear memory system from the start, or it will drift as tasks pile up. If you expect the work to grow, plan for separate specialist agents and keep a human check before any message is sent or change is made.

Try it yourself with Codex

Help me design an always-on agent for [company or team] that handles [recurring tasks]. Run it as a [systemd service / scheduled job] on [machine or environment], and have it read work from [email inbox / folder / queue]. Keep its memory in [daily diary file], [markdown wiki folder], and a rolling context summary. First, make a plan and wait for my approval before doing anything that sends email, deletes files, submits forms, or changes important data. Then carry out one test task from start to finish and report exactly what happened.

Fill in the [brackets]. Agents can make mistakes, so review anything before it is sent, paid or deleted.

Read the original on news.ycombinator.com โ†—

Get 3 agent uses like this every morning

With the day's top AI news. Free, and one click to unsubscribe.