Superhuman Go Wants to Be the Agent That Bosses Other Agents
Superhuman, the company that used to be Grammarly, is pushing its Go assistant into everything it owns. TechCrunch has the rollout (https://techcrunch.com/2026/11/19/superhuman-pushes-its-go-assistant-everywhere/, November 19, 14:00 UTC), and CEO Shishir Mehrotra put a number on it: Go has more than 1 million active users. Product: https://superhuman.com
What changed: Go now learns context across connected apps, and onboarding asks you to hook up your tools and load work context up front. It works inline at the cursor, the old Grammarly trick, and it also manages other agents. It goes proactive, watching your calendar and inbox and drafting replies in your writing style. Twenty new connected agents arrive through the Go SDK, including DeepL, Legora, Amplitude, Databricks and Blackboard. In Docs, every Go agent runs natively, can be sent off on background tasks, and can be wired into routines that fire on events. In Superhuman Mail, Go replaces the AI that was already there, and does enterprise search over the company knowledge base. Fathom, bought in September, now powers shared live notetaking, and Go drafts things like a first proposal after the meeting.
The interesting bet is the shape, not the feature list. Superhuman is not trying to have the smartest model. It wants to be the layer that already sits in your cursor, your inbox and your docs, and that other vendors' agents plug into. Mehrotra says the target is OpenAI's Dots and Claude Duet more than consumer agents like Muse or Instinct. That is a distribution play: Grammarly spent the better part of two decades getting installed everywhere people type, and Go is the attempt to cash that in before the model labs reach the same surfaces.
The catch is the same one Meta hit this week. A proactive agent that reads email and calendar, drafts in your voice and calls 20 third-party agents is a lot of reach for one permission screen. One million active users is real traction. The next number worth asking for is how many of those drafts get sent without edits, and how many get sent at all.
← Back to all articles
What changed: Go now learns context across connected apps, and onboarding asks you to hook up your tools and load work context up front. It works inline at the cursor, the old Grammarly trick, and it also manages other agents. It goes proactive, watching your calendar and inbox and drafting replies in your writing style. Twenty new connected agents arrive through the Go SDK, including DeepL, Legora, Amplitude, Databricks and Blackboard. In Docs, every Go agent runs natively, can be sent off on background tasks, and can be wired into routines that fire on events. In Superhuman Mail, Go replaces the AI that was already there, and does enterprise search over the company knowledge base. Fathom, bought in September, now powers shared live notetaking, and Go drafts things like a first proposal after the meeting.
The interesting bet is the shape, not the feature list. Superhuman is not trying to have the smartest model. It wants to be the layer that already sits in your cursor, your inbox and your docs, and that other vendors' agents plug into. Mehrotra says the target is OpenAI's Dots and Claude Duet more than consumer agents like Muse or Instinct. That is a distribution play: Grammarly spent the better part of two decades getting installed everywhere people type, and Go is the attempt to cash that in before the model labs reach the same surfaces.
The catch is the same one Meta hit this week. A proactive agent that reads email and calendar, drafts in your voice and calls 20 third-party agents is a lot of reach for one permission screen. One million active users is real traction. The next number worth asking for is how many of those drafts get sent without edits, and how many get sent at all.
Comments