Search “AI to-do list” and the first page is apps that do one thing with the AI: reorder the list. Priority, time blocking, natural-language entry. The list is smarter. It is still your list to do.
This post is about the other thing. A to-do list where you add an item and it gets done.
What “does the task” means
A concrete test. Add these five items and see how many are closed by the tool without you touching them:
- Reply to the three unanswered support emails from this week.
- Fix the failing test in
auth/session.test.tsand open a PR. - Write the September changelog from the merged PRs and post it to the blog.
- Pull last month’s Stripe revenue into the investor update sheet.
- Reply to the five people who tagged us on X yesterday.
An organizer scores zero: it can schedule the five into your calendar. Zapier-style automation closes only the items you built a flow for beforehand. A chat assistant drafts. ChatGPT Work goes further and hands back finished sheets and docs, but its browser keeps its own sessions, so your support inbox and X account need a separate sign-in there.
For the tool to score five it needs, at minimum:
- Your logins. Support inbox, GitHub, the blog CMS, Stripe, X. Either a CLI or API credential it can use, or your actual browser session where there is no API worth using.
- A machine. Somewhere to clone the repo, run the tests, keep the files between tasks.
- A browser it can drive. Stripe and GitHub have good APIs; the ad manager’s bulk editor, most CMS admins and the supplier portal do not.
- Permission to act, with a way for you to say “ask first” on the risky ones.
That is the checklist. Everything else on the page is how TODO for AI meets it, and where it does not.
How TODO for AI does it
The list is the interface: each TODO becomes a task for an agent, with its own chat where it reports back. What it can reach depends on what you connect:
- Your browser. The Chrome extension works in the tabs you attach and the tabs it opens, so it inherits your Gmail, X, Stripe and CMS sessions. How that works and what it can reach is in its own post.
- A machine. A cloud VM per account, or your own computer via the bridge, so it can clone the repo and run the tests.
- Any model. Claude, GPT, Gemini, Grok or Ollama per task, on one plan. The free plan is Sonnet and Haiku; the full catalogue starts at $20.
- Tool permissions. Per tool: allow, ask first, or block. Any run can be stopped. Secrets live in an encrypted vault.
Against the five-item test: the extension covers 1, 3 and 5 while Chrome is running; a paired machine or the cloud VM with the repo covers 2; 4 needs Stripe open in an attached tab, or the Stripe CLI logged in from the vault, in which case it runs on the VM with nothing of yours open.
Where it stops
It is not zero setup. You install the extension, attach the tabs it may use, and pair a machine if you want it in a repo. That is the price of the agent working where you work instead of in a sandbox with nothing in it.
Browser items need your browser open. Anything that runs through the extension runs in your Chrome, so “close the laptop and it keeps going” only holds for work on the cloud VM: repos, CLIs, APIs, and research in the cloud browser. A queue of inbox and ad-account items waits for Chrome.
“Ask first” is a tool-level gate, not a policy engine. It stops before a whole tool — a shell command, a browser action — not before “a refund over $200”. Nothing in the product understands that one stripe command charges a card and the next one reads a balance. If money is involved, set that tool to ask first, or write “draft and wait for me” in the TODO, and check the diff before you approve.
Proposals need a source. It proposes TODOs from your business once it has something to read: your site, an inbox it has a tab for, a repo it is paired with. Nothing is proposed from a source it has not been connected to.
Compared to what ranks for “AI to-do list”
| Todoist, Motion | Lindy | ChatGPT Work | TODO for AI | |
|---|---|---|---|---|
| Sorts and schedules the list | Yes | No | No | Yes |
| Does the items | No | The ones you built a workflow for | Deliverables: sheets, docs, slides | Yes |
| Uses your logged-in browser | No | No | No, its own | Yes, attached tabs via Chrome extension |
| Runs on your machine or a persistent VM | No | Cloud only | Its cloud, local files via desktop app | Both |
| Model choice | n/a | Most major models | OpenAI only | Every vendor, one plan |
| From | Free | $29.99/mo | Plus $20/mo | Free on Sonnet, $20/mo for every model |
Prices checked 2026-09-24; the ones that move are on the comparison pages.
Todoist and Motion are good at what they do. If your problem is knowing what to do next, buy one of them. If your problem is that the list is correct and still nothing on it gets done, that is the problem this was built for. Which agent fits which job is scored on the founders guide.
Try the five-item test
todofor.ai is free to start, no signup, $2 of usage included. Paste your own five items. If it closes fewer than three, tell us which ones and why; that is the roadmap.