The comparison

Every AI agent, one table

15 tools, 5 questions, no marketing. Work platforms are adding agents, agent vendors are adding task lists, and neither has arrived. Here is exactly who has what.

Start free Pricing
15
tools compared
5
questions each
1
row without a hole
2026-09-22
verified
The map

Two groups walking toward the same product

Linear, Notion, ClickUp and Asana already had the thing agent vendors do not: a board where a colleague and you hold the same item, with state everyone can see. In the last year all four put agents on it. Notion Custom Agents pick their own model and run on triggers. ClickUp Super Agents get @mentioned in comments. Linear lets you assign an issue to Claude Code, Codex, Cursor or Devin.

Coming the other way, Claude Code, Cowork, Codex, ChatGPT, Manus, OpenClaw and Hermes all shipped some form of task list. Every one of them is private. It is your queue, rendered; there is no version where a teammate opens it and sees what the agent is doing. GitHub is the exception, because issues were already a shared board before Copilot arrived on them.

Neither convoy has arrived. The work platforms have the board and their agents cannot run a command. The agent vendors have the worker and no shared place to put the work. GitHub has both and only accepts work shaped like a pull request.

The test

A row with no holes

ToolShared boardOpen-ended workShell you ownYour browserModel choice
Work platforms that added agents
LinearShared board. Yes: Yes, issuesOpen-ended work. No: Engineering onlyShell you own. Partly: The delegated agent'sYour browser. No: NoModel choice. No: The delegated agent's
NotionShared board. Yes: Yes, agents shareableOpen-ended work. Partly: Workspace and MCP appsShell you own. No: NoYour browser. Partly: Its own browsingModel choice. Yes: Claude, GPT, Gemini, Grok
ClickUpShared board. Yes: Yes, assignableOpen-ended work. Partly: Workspace and skillsShell you own. No: NoYour browser. No: NoModel choice. Yes: Max, Claude, GPT, Gemini
AsanaShared board. Yes: Yes, AI teammatesOpen-ended work. No: Build the flow firstShell you own. No: NoYour browser. No: NoModel choice. No: Vendor managed
LindyShared board. Partly: Per user, shared poolOpen-ended work. No: Build the flow firstShell you own. No: NoYour browser. No: NoModel choice. Yes: Per node
Agent vendors that added a task list
Claude CoworkShared board. No: Private tasksOpen-ended work. Yes: YesShell you own. Yes: Your foldersYour browser. Yes: Claude in ChromeModel choice. No: Anthropic only
Claude CodeShared board. No: Private sessionOpen-ended work. Partly: Repo first, MCP beyondShell you own. Yes: YesYour browser. Yes: Claude in ChromeModel choice. No: Anthropic only
OpenAI CodexShared board. No: Private task listOpen-ended work. Partly: Repo shapedShell you own. Yes: YesYour browser. No: NoModel choice. No: OpenAI only
CursorShared board. No: Private, repo configOpen-ended work. Partly: Repo shapedShell you own. Yes: YesYour browser. Partly: To check your appModel choice. Yes: Multi-vendor, metered
GitHub CopilotShared board. Yes: Yes, GitHub issuesOpen-ended work. Partly: Repo shapedShell you own. Yes: CLI local, cloud on runnersYour browser. No: NoModel choice. Yes: OpenAI, Anthropic, Google, xAI
ChatGPTShared board. No: Private chatsOpen-ended work. Yes: YesShell you own. No: Their sandboxYour browser. Partly: Its own, plus connectorsModel choice. No: OpenAI only
ManusShared board. No: Tasks shared one by oneOpen-ended work. Yes: YesShell you own. No: Their cloudYour browser. Yes: Browser Operator extensionModel choice. No: Vendor picks
OpenClawShared board. No: Chat threadsOpen-ended work. Yes: YesShell you own. Yes: The box you hostYour browser. Yes: CDP, self-configuredModel choice. Yes: Any, BYOK
Hermes AgentShared board. No: Chat threadsOpen-ended work. Yes: YesShell you own. Yes: The box you hostYour browser. Yes: CDP, self-configuredModel choice. Yes: Any, BYOK
TODO for AIShared board. Yes: Yes, one boardOpen-ended work. Yes: YesShell you own. Yes: Your PC + cloud VMYour browser. Yes: Your Chrome profileModel choice. Yes: Every provider, per agent
Our take
Every column here is matched by somebody. Notion and ClickUp let you pick the model, GitHub has a shared board and a local shell, Cowork acts in your Chrome. One row has all 5 cells filled, and the closest miss is GitHub, which only accepts work that can become a pull request.
The gap

Where the two columns cross

Read the table down the board column and the shell column together. Linear, Notion, ClickUp and Asana have the board, and their agents never get a command line. Cowork, Claude Code, Codex, Cursor, OpenClaw and Hermes have the command line, and the work stays in a private queue.

That is the whole difference. A Notion agent can update a database, file a bug and post a report, and it cannot run your deploy script. Claude Cowork can run your deploy script, and your colleague cannot see that it did.

GitHub is the one product on both sides of that line: issues are a real shared board and Copilot CLI runs on your machine. Its limit is shape. Work has to become a repository, an issue and a diff, so the invoice, the ad account and the vendor portal never enter. Ours is the same combination without the repository in the middle, plus a session in your own browser.

It matters because most real work is mixed. The same week contains a code fix, an invoice to chase, a landing page to ship and a vendor portal to log into. Put that on a work platform and half of it stalls at the connector boundary. Put it in a chat agent and nobody else knows it exists.

01
A board, not a thread

Items have state: open, running, blocked on you, done. Contributors and the agent that ran each item are recorded, so a colleague can open the board and read what happened without you narrating it.

02
An agent per item, not one assistant

Each todo names its own agent, and each agent has its own model, tools, system prompt and permissions. A code fix and an invoice chase can sit on the same board without one badly-configured assistant doing both.

03
A shell you own, on purpose

The 90 KB bridge attaches your own machines, and every account gets a managed cloud VM that keeps files, logins and sessions between runs. Any binary you have installed is a tool, no MCP server required.

04
Permissions in front of the shell

Every tool call resolves against an allow, ask or deny rule, most specific match first, configured per agent. It gates which commands run, so you can hand one agent the whole shell and another only the two binaries it needs. It is a policy layer, not an OS sandbox: whatever you allow runs with your own user rights.

Honestly

Where the others are ahead

Model choice is not our moat. Cursor, Copilot, Notion, ClickUp, Lindy, OpenClaw and Hermes all let you pick the vendor. It only separates us from the agents that are a front end for one lab.

Local execution is not our moat either. Cowork works on your folders, Claude Code, Codex and Copilot CLI live in your terminal, and OpenClaw and Hermes run on a box you host. Anyone who tells you their agent is the only one that runs on your computer is selling.

And per product: Linear is a better issue tracker than our board, Notion is a better knowledge base than anything we have, Asana has governance we do not, GitHub has distribution nobody can buy, and Claude Code is a stronger pure coding agent than most. What none of them has is the whole row.

FAQ

Common questions

Is this table maintained?
The names and logos come from the same competitor data that drives every /vs page; the five cells are judged by hand against vendor documentation, last verified 2026-09-22. When a vendor ships something that fills one of their holes, the cell changes. Several already did: Notion agents pick their own model, Cursor and Copilot are multi-vendor, Cowork works on local folders.
What counts as a shared board?
A place where a person and an agent can hold the same item, where the item has state the whole team can see, and where you can tell afterwards who and what worked on it. A private task list inside a chat product does not count, and neither does a credit pool or a project whose tasks are shared one at a time.
What counts as a shell you own?
A command line on a machine you control: your laptop, your server, or a VM assigned to your account that keeps state between runs. A vendor sandbox that resets per task does not count, because your logins, your tools and your files are not in it.
Why is model choice a column if it is not a differentiator?
Because it still separates us from every agent that is a front end for one lab, and because buyers ask. We would rather show a column that most competitors match than leave it out and look like we are hiding the comparison.
Can I use TODO for AI alongside these tools?
Most people do. Keep Linear for engineering issues, Notion for knowledge, Cursor for typing, and put the work that needs a machine, a browser session or several kinds of expertise on a TODO board.