How to use Notion to generate and edit any file (Notion AI computer access)
While working on a PowerPoint presentation for an upcoming investor meeting, John—the CEO of a fast-growing U.K. scale-up—let a Notion Custom Agent use all the notes in the company’s Notion workspace to craft the presentation. Minutes later, the deck was ready for review.
Fantasy or reality? More reality than fantasy, since AI sometimes crashes or produces outputs far below your desires; and at the same time the story above is real and feasible.
We call the ability of Notion AI to manipulate, analyze, or generate any file (and more) computer access.
Computer access allows Notion Custom Agents to create files (for example, PDF, PPT, and Word documents) that you can easily download, among other things. This is part of the agent‑orchestration paradigm I have explored previously.
Computer access means that AI can use an isolated Linux sandbox (a remote virtual machine) to run code and generate artifacts. This includes creating files of any kind, doing data analysis, writing code, and more. It doesn’t have access to your local files—only to the files it generates, plus your Notion workspace and any third‑party connections you add.
In Notion, this capability is inherent to both Notion AI Agent (the chat always available at the bottom-right corner) and Notion Custom Agents (running on a schedule or specific triggers), so you don’t have to set up anything.
You can chat with the agent and request what you want done, or let it run automatically based on triggers or a schedule; computer tools are invoked whenever suitable. I covered Notion Custom Agents in detail in this essay and this video.
Notion AI using Computer tools call
What it can do:
Run code — “Take this file, run data analysis on it, and save the result here.”
Create files — for example, generate a .pptx, .xlsx, .pdf, charts, or images
Process files you upload into the chat (read, parse, edit)
Do data analysis
Let you download any file it generates
Use MCP tools
What it can't do:
See or touch your local Mac or PC (no access to your Downloads, Desktop, and so on)
Control your browser or apps on your machine
The friction of keeping data updated across systems has been an important blocker in software tool adoption. Agentic artificial intelligence is bridging that gap, allowing you to manage and do work in the same surface: a place where writing and maintaining documentation becomes increasingly valuable.
Those documents feed the agents the context they need, and the agents feed them back with updates. The first 60% of John’s work was done in minutes, and John could now refine it to completion.
