The AI everyone goes to, or the AI that reads your messages.
ChatGPT is OpenAI's general-purpose AI assistant. You open it and ask it to think, write, code, delegate a browser task to ChatGPT Agent, or pull content in through Apps. this+that is different. It's a comms-forward AI that reads every email, Slack thread, Teams chat, Google Chat, and Telegram message you receive, extracts the tasks, and runs workflows. Both products speak MCP. Many teams use both.
A general-purpose AI you open and ask, or a comms-forward AI that reads your messages
Software has been here before. In the early Office days, Access was a core component because people built their own line-of-business apps. Then specialized SaaS took over, one domain at a time. Salesforce won CRM. Asana won projects. Shopify won commerce. Even though some people use Excel as a database, Access did not lose to Excel. It lost to a constellation of purpose-built apps that each nailed one job better than a general tool could. AI is at the same point. We think of it as two waves: Wave 1 is the general-purpose chat you open and ask; Wave 2 is the fit-for-purpose AI that owns one job end-to-end, with UI tailored to it (more in our essay on AI's second wave).
ChatGPT is a Wave 1 product. Hundreds of millions of users, real Memory, Projects with their own scoped context, Custom GPTs, ChatGPT Agent and Atlas for browser delegation, Tasks for scheduled prompts, Apps for content connectors. OpenAI is clearly walking toward the Wave 2 frame. Even so, the product still requires you to open ChatGPT and ask. Tasks cap at ten. Agent runs when invoked. Memory is about your chats, not your inbox. this+that is one of Wave 2's first products, fit-for-purpose for messages, tasks, and team workflow. The agent lives inside your inbox and chat channels, scans every message for what needs doing, surfaces tasks for the whole team, and runs workflows automatically. Both products are MCP-native, so they compose. ChatGPT can reach this+that's DoBox through MCP. Most teams who use ChatGPT heavily will also benefit from a Wave 2 product handling their inbox. They solve different jobs.
Feature by feature
How the two products compare across the things that matter most.
| Topic | ChatGPT | this+that |
|---|---|---|
| Category | Wave 1 general-purpose AI. A chat product you open and ask, with Memory, Projects, Tasks, Agent, and Apps stacked around the conversation. | Wave 2 comms-forward AI. An agent that reads your messages and acts on them without being asked. |
| Where work starts | With you. You prompt ChatGPT in a chat, run Agent on a browser task, schedule a Task to fire later, or open an App to pull content in. | With incoming messages. The agent reads what arrives without being asked and surfaces what needs doing. |
| Channels read | None natively. Apps (Gmail, Outlook, Slack, Teams, Drive, SharePoint, GitHub, HubSpot, Asana, Box, Dropbox) pull content in when you ask. Not continuous scanning. Apps are off by default for Enterprise and Edu admins. | Gmail, Outlook, Slack, Teams, Google Chat, and Telegram, continuously. Instagram and Facebook Messenger awaiting approval. |
| Task layer | Tasks lets you schedule a prompt to run later (web, iOS, Android, macOS). Hard cap of ten active tasks per user across every plan. Project Memory keeps notes scoped to a workspace. | DoBox. Every message scanned for tasks, surfaced as assignable items for the whole team, with no per-user cap. |
| Workflow automation | ChatGPT Agent runs browser, terminal, and file tasks with per-action approval when you invoke it. Also reachable through the Atlas browser. Workspace Agents replace Custom GPTs for org-wide flows in Slack, Salesforce, and connected apps. | Plain-language workflow builder triggered by incoming messages. 32 action types with IF/ELSE branching, loops, and sub-agents. |
| Memory / knowledge | Saved Memories (an editable list of explicit facts) plus Reference Chat History (implicit pattern recall). Project Memory is scoped per workspace and does not bleed across projects. Dreaming, a background-synthesis architecture, began rolling out in June 2026. Single-user across all of it. | The brain. A writable team and personal knowledge layer with versioning. Agents read and write as a side effect of doing real work. |
| Team capabilities | Business plan ($20/seat/mo annual, $25-30/seat/mo monthly, 2-seat minimum) with shared admin, SSO, and connectors. Memory and conversations are still individual. Enterprise adds SCIM, audit logs, and per-role controls. | Team task assignment, shared task view, and team-request grounding (your agent searches a teammate's brain to draft a suggested reply). |
| MCP support | Native. Developer Mode connects ChatGPT to remote MCP servers; the Responses API and Apps SDK both build on MCP. | Native, bidirectional. Consumes any MCP server in workflows (HubSpot, GitHub, Linear, etc.), and exposes a public MCP server so ChatGPT, Claude Desktop, or any MCP client can reach your tasks. |
| Integrations | Apps catalog of OpenAI-vetted connectors, plus any tool via MCP. | Open MCP standard. Any tool with an MCP server or API plugs in. |
| Surfaces | Web, macOS, Windows, iOS, Android, the Atlas browser, and the API. | Web app, browser extension (DoBox for Gmail), and a public MCP server reachable from any AI client. |
| Pricing | Free, Go $8/mo, Plus $20/mo, Pro $200/mo, Business $20/seat/mo annual ($25-30/seat/mo monthly, 2-seat min), Enterprise custom. Verify current tiers on chatgpt.com/pricing. | Free during beta. |
- ChatGPT is the most widely used general-purpose AI. For reasoning, writing, analysis, and code, the model and the surrounding product are excellent and improving fast.
- ChatGPT Agent (and the Atlas browser) handle delegated browser, terminal, and file tasks with per-action approval. Strong personal agent for the work you ask it to do.
- Memory is real: Saved Memories, Reference Chat History, Project Memory scoped per workspace, and the Dreaming background-synthesis rollout that started in June 2026.
- Tasks let you schedule prompts to fire later. Useful for recurring personal briefings, with a ten-task per-user cap.
- Apps connect ChatGPT to Gmail, Slack, Outlook, Teams, Drive, SharePoint, GitHub, HubSpot, Asana, and more so you can pull content into chat without copy-paste.
- Business at $20/seat (annual) ships with SSO, admin, and connectors. Enterprise adds SCIM, audit logs, and granular access controls.
- A different category. The agent reads the messages you receive, not just the prompts you send.
- Reads six channels natively, continuously: Gmail, Outlook, Slack, Teams, Google Chat, and Telegram. Not pull-on-demand, ambient. Instagram and Facebook Messenger awaiting approval.
- DoBox: every message scanned for tasks, with the whole team able to see, assign, and track them in one place. No ten-task cap.
- Workflows triggered by incoming messages, not just by you asking or by a schedule. 32 action types with branching, loops, and sub-agents.
- The brain is a writable team knowledge layer (not single-user Memory) that agents read and update as a side effect of doing real work.
- MCP works both ways: this+that consumes any MCP server inside workflows, and exposes its own so ChatGPT, Claude Desktop, or any MCP client can reach your tasks. The two products compose rather than compete.
See what's waiting in your inbox
Connect your inbox and get a free analysis of the tasks, follow-ups, and open commitments this+that would surface across your channels.
Free to use · No credit card required