Deep Research and Agent Mode
People who need researched, cited deliverables and multi-step tasks done, and want to delegate them rather than babysit them
Prerequisites: Module 303 (ChatGPT Projects) and Module 304 (Connectors and Apps) or equivalent experience
Last verified 2026-09-17
Learning objectives
By the end, participants can:
- Start a deep research task (type @Deepresearch in the composer, or pick Deep research from the tools (+) menu in Chat) with a brief that names outcome, audience, constraints and sources
- Review a research report the way a sceptical editor would: claims, citations, disagreements
- Explain what ChatGPT Work is, what agent mode was (retired; ChatGPT Work, with its cloud browser, is the successor for multi-step tasks), and where each runs
- Delegate a task to Work with the cloud browser (ChatGPT Work on paid plans, excluding Free and Go) and handle sign-in and confirmations safely
- Draw an autonomy line: which actions run unattended, which pause, which never happen through an agent
What's in this module
7 lessons · 5 exercises · ~7 min reading
Starts withTwo kinds of delegation~1 minExecutive summary
You will learn how to brief and review a deep research task, and how to delegate a multi-step task to ChatGPT Work, the agent that replaced agent mode, including its cloud browser, its confirmations, and the line between what it may do alone and what stays with you. End state is one deep research report you have reviewed source by source, one Work task that ran in the cloud and paused for your approval, and a written autonomy line for your own use. After this module you can write a research brief that gets a usable report, tell a good report from a fluent one, delegate a browser task without handing over your passwords, and decide what an agent may do without asking.
Sources
- OpenAI Help Center, "Deep research in ChatGPT", how to start a task, how the plan and sources work, and usage by plan: https://help.openai.com/en/articles/10500283-deep-research-in-chatgpt
- OpenAI Help Center, "ChatGPT Work and Codex", what Work is, where it runs, cloud versus local, and workspace controls: https://help.openai.com/en/articles/20001275-chatgpt-work-and-codex
- OpenAI Help Center, "Using cloud browser in ChatGPT", website permissions, the secure sign-in flow, consequential-action confirmations and take-over: https://help.openai.com/en/articles/20001280-using-cloud-browser-in-chatgpt
- OpenAI Help Center, "ChatGPT agent", the page that now records agent mode's retirement and points to Work: https://help.openai.com/en/articles/11752874-chatgpt-agent
- OpenAI Help Center, "Scheduled tasks in ChatGPT", recurring and event-triggered tasks and their limits: https://help.openai.com/en/articles/10291617-scheduled-tasks-in-chatgpt
Prefer the live room?
This module also runs inside our in-person bootcamps and workshops in Copenhagen.
More in Track D
Claude for Excel
How to get real multi-tab spreadsheets with working formulas out of Cowork, with you controlling the assumptions.
AI Slides
The research-first pipeline that produces decks worth presenting, plus the brand setup that keeps a whole team on-style.
Claude Design
What Claude Design produces (sites, decks, video) and the design-system workflow that keeps output on-brand instead of generic.