Hand Codex work orders through a Dot that keeps the worker moving

The poster builds small Windows tools for their own business. They planned the work in regular ChatGPT and had Codex do the coding, but kept having to relay messages and nudge Codex along by hand. After moving to the $100 Pro plan they set up a Dot, OpenAI's always-on agent, as the middle layer, talking to them through Slack. ChatGPT writes a work order, the Dot picks it up and hands it to Codex, watches the run, and messages the poster in Slack once the job is finished or when it hits a choice that needs them.

One day in, the poster says the Dot took a lot of busywork off their plate. Reading the exchange between the Dot and Codex, they saw it tell Codex that a stop was not a real hard gate and to split the job into smaller parts, that the objective was not done yet, and to try another route when the environment blocked a command. Those were nudges the poster used to give by hand. No figures on time or credits were given.

Commenters were mixed. One says the Dot micromanages Codex and that the loop takes longer and burns more credits than giving Codex one prompt. Another reports it getting stuck on permission prompts it could not clear. One user has a Dot overseeing a four-day Codex job with updates every two hours. A commenter suggests giving the Dot a short list of things it must escalate rather than push past: deleting files, credentials, payments, pushing to main.