Technology

Superhuman Brings Its Go Agent Into Docs, With Background Automation and Wider Integrations

Martin HollowayPublished 8m ago3 min readBased on 1 source
Reading level
Superhuman Brings Its Go Agent Into Docs, With Background Automation and Wider Integrations
source:superhuman.com

Superhuman, formerly known as Grammarly, has made its Go assistant available natively in Docs, and CEO Shishir Mehrotra said Go has over 1 million active users. The company's announcement was reported by TechCrunch on November 19, 2026.

Go learns each user's writing style and adjusts its responses in email and Docs. It uses surrounding context to build drafts and to organize the inbox across both Google and Outlook accounts. Inside Docs, Go works inline at the cursor, and it can also manage other agents. Here, an agent means software that carries out a task on the user's behalf rather than answering a single prompt.

The Docs integration adds two automation features. Users can set Go agents to work on different tasks in the background. They can also configure routines, which trigger workflows when a specified event occurs.

In Superhuman Mail, Go is replacing AI as the central tool, according to the report. The change moves the product's branding and interaction model around a single agent rather than a set of separate AI features.

Superhuman has also widened the set of third-party agents that plug into Go. Through the Go SDK, a software development kit for building integrations, the company added 20 new connected agents, including DeepL, Legora, Amplitude, Databricks, and Blackboard. The list spans translation, legal work, product analytics, data platforms, and learning management, which indicates how far the assistant is intended to reach beyond email and documents.

On the context side, Superhuman acquired the notetaker Fathom in September to add meeting context to its suite. The company is now rolling out live collaborative notetaking that is shared across the team.

The broader context here is that Superhuman is building its agent layer on context rather than on generation alone. A drafting model that sees only the open message has limited information. One that can draw on meeting notes, inbox history, and connected tools has a richer basis for action, and the Fathom acquisition and the SDK partners both point toward that goal. The same design widens the surface an agent can touch, so the permission model for background agents and event-triggered routines becomes central to how the product is used.

In my view, the most useful question for teams evaluating this kind of assistant is not whether the drafts read well. It is which systems the agent can read and write, under whose credentials, and how a routine behaves when the event it listens for fires at an unexpected time. The announcement does not answer those operational questions in detail.

The optimistic reading is also a real one. An assistant that keeps a team's meeting notes, mail threads, and document edits in one place could remove a good deal of manual coordination, and the 1 million active users Mehrotra cited suggest a sizable group already finds the approach worth using. Whether that scale turns into lasting gains in how teams work will depend on how carefully the new automation is governed.