Deep Research, Apps, and Agentic Work¶
Deep research¶
One of ChatGPT's biggest shifts has been from "answer my question" toward "go and do a substantial research task for me." OpenAI's deep research feature is the clearest example. It is designed for complex online tasks: the system researches, reasons, and synthesises material into a documented report, working from public web sources, uploaded files, specific websites, and enabled apps while keeping the user in control.
For humanities use, deep research is potentially valuable for:
- Exploratory literature overviews
- Policy or funding landscape scans
- Background briefings on current debates
- Preliminary comparisons of institutions or archives
- Synthesis across a large number of contemporary web sources
It is much less suitable as a substitute for primary-source scholarship, and it should not be treated as a reliable bibliographic authority without careful checking.
Leif's Notes
Deep research is a research assistant for scoping; used badly, it becomes a polished source of false confidence. The reports look authoritative. That is precisely when you need to check the sources most carefully. See Verification & Citation for how to do this systematically.
Apps¶
What OpenAI previously called connectors is now generally referred to as apps, though both terms still appear in their documentation. Apps can do two rather different things:
- Some apps add interactive interfaces inside ChatGPT.
- Others let ChatGPT search and reference information from connected third-party services such as Google Drive, SharePoint, GitHub, Box, and other tools.
Some apps can also be used inside deep research. Availability varies by plan, app, and region.
Agent mode¶
OpenAI also offers ChatGPT agent mode, which is more interventionist than deep research. Agent mode can navigate websites, work with uploaded files, connect to third-party data sources, fill out forms, edit spreadsheets, and carry out complex online tasks while keeping the user in control. (Agent mode, launched in July 2025, integrates the capabilities of the earlier standalone Operator product.)
Agent mode can also schedule recurring tasks — daily, weekly, or monthly — which is useful for monitoring or routine data collection.
For most humanities researchers, agent mode is not the starting point. It is for moments when you genuinely want delegated action, not just delegated thinking.
A practical sequence¶
A sensible rule is:
- Use ordinary chat first.
- Use deep research when you need broad web-based synthesis.
- Use apps when you need access to your own tools or repositories.
- Use agent mode only when the task truly requires action as well as analysis.