Give Your dot Ongoing Work: Goals, Schedules and Triggers
The real power of a dot is that it keeps going when you stop. This lesson shows the three ways to give it ongoing work: a goal it works toward, a schedule it follows, and triggers it watches for. You will also learn how to write instructions that a background agent can follow without you there.
What You'll Learn
- The difference between a goal, a schedule and a trigger
- How your dot's cloud computer and browser work, and when you need to take over
- How memory works for a dot
- How to write a clear brief for work that runs while you are away
- How bigger jobs like Codex tasks fit in
Three Kinds of Ongoing Work
| Type | What it means | Example |
|---|---|---|
| Goal | An outcome it works toward over days | "Plan the team offsite in March" |
| Schedule | Something it does at set times | "Every Friday at 4pm, sum up the week's open threads" |
| Trigger | Something it reacts to | "Tell me when someone in #support mentions billing" |
Most useful setups mix all three. A goal like "plan the offsite" might include a schedule ("send me a status every Monday") and a trigger ("tell me when the venue replies").
Goals
When you give a dot a goal, it breaks it into steps and keeps working in the background. It can run several tasks at once, and it can pause and pick up again on its own. It comes back to you with results, or with a question when it needs your judgment.
Schedules
You can ask for reminders, recurring checks and scheduled updates. Your dot keeps them under Scheduled, where you can see, change or delete them. Use plain language:
Every Monday at 9am, check the shared "Offsite" folder and tell me
what changed last week, what is still undecided, and what is due
this week.
Triggers
Your dot only watches the things you name. In Slack, adding it to a channel is not enough. You have to tell it which channel and what to look for:
Watch #customer-feedback. When three or more people mention the
same problem in one week, send me a short summary with links to
the messages. Do not reply in the channel.
Its Own Cloud Computer and Browser
Every dot has its own computer and browser in the cloud. That is where it opens websites, reads pages and fills in forms for you. It is separate from your own browser: it does not share your cookies or saved logins.
Sometimes a site needs you to sign in. Your dot will ask, and you can take over the browser, sign in yourself, then hand control back. When you share a password, it goes through a private form straight to the cloud browser, and OpenAI says the model does not see it. Reusing a saved login later needs your OK.
Two practical tips:
- Expect a short delay when it starts the cloud computer. It is built for background work, not instant answers.
- If a site keeps logging it out, do that task yourself. Not every website works well with agents.
How Memory Works
Your dot reads your normal ChatGPT memory, and it also keeps its own notes as it works. Those notes stay until you delete the dot. So:
- Tell it your preferences once ("I prefer short summaries with bullet points") and it should remember.
- Do not share things you would not want kept, like full ID numbers or passwords in chat.
- If it remembers something wrong, tell it to correct the note.
Write a Brief That Works Without You
A chat can be vague because you are there to clarify. A background agent cannot ask you every few seconds. Good briefs answer five questions:
- Goal. What does "done" look like?
- Sources. Which apps, folders or channels should it use?
- Timing. When should it check, and when should it report?
- Limits. What must it never do without asking?
- Format. How should the report look?
Here is a full brief using all five:
Goal: Help me get ready for my job interviews this month.
Sources: My Gmail label "Applications" and my calendar.
Timing: Check every evening at 7pm. Report only if something changed.
Limits: Never reply to recruiters or accept invites. Draft only.
Format: A short list: new messages, upcoming interviews with date,
time and company, and one thing I should prepare for each.
Compare that with "help me with my job search". The second one leaves the dot guessing, and a guessing agent wastes your usage and your time.
Bigger Jobs: Work and Codex Tasks
A dot can hand bigger jobs to other ChatGPT tools. It can start Work tasks, Codex tasks (for code) and Codex cloud tasks, and it can run local skills if you connected your computer.
Keep one cost rule in mind. On Pro, chatting with your dot does not count toward your ChatGPT limits. But the Work and Codex tasks it starts do count toward those products' limits. If you see limits drop fast, check what your dot has been starting.
Check In, Then Step Back
For the first week, check in once a day:
- Ask: "What did you do today, and what is waiting for me?"
- Look at Scheduled and delete anything you no longer need.
- Correct anything it got wrong, so its notes improve.
After that, most people only need to read its updates.
Key Takeaways
- Give a dot ongoing work as a goal, a schedule or a trigger, or a mix.
- Schedules live under Scheduled. Triggers only fire for the channels and events you name.
- Your dot has its own cloud computer and browser, separate from yours. You can take over to sign in.
- It reads ChatGPT memory and keeps its own notes until you delete the dot.
- A good brief covers goal, sources, timing, limits and format.
- Work and Codex tasks your dot starts count toward those products' limits.
Quiz
Up next · Lesson 4 of 5
Custom Rules and Safety: What Your dot May Do Alone
4 min read
3 of 5 lessons done

Ask Byte about this lesson
A private chat about this lesson. Byte has read the whole lesson, answers in seconds, and shows the section each answer comes from. Ask follow-ups in any language.
Questions & Answers
Public questions about this lesson. The AI Tutor answers right away in your language, the FreeAcademy team checks answers, and every learner can read them.

