Why Grammarly Could Buy Superhuman When Everyone Thought It Was the One in Trouble

When the deal dropped in late June — Grammarly buying Superhuman — the first reaction in a lot of founder group chats was the same: wait, isn’t Grammarly the one getting eaten by AI? Didn’t ChatGPT and Claude make “fix my writing” a commodity? How can the allegedly disrupted company buy the allegedly thriving AI email startup?

The short version is that the narrative was backwards. Grammarly never actually collapsed; it broadened. Superhuman never actually became a mass-market rocket; it became a premium, LLM-augmented surface with great users but bounded TAM. Put those together and the 2025 transaction suddenly looks obvious. And once Grammarly rebranded the whole company to Superhuman in October, folding in Coda and the new Superhuman Go agent layer, it became even clearer what they were doing: reassemble multiple high-frequency writing surfaces under one AI productivity brand. Sources: TechCrunch, July 1, 2025, Grammarly company blog, June 30, 2025, Reuters, The Verge, Oct 29, 2025

The rumor vs. the P&L

The rumor was that Grammarly was “past its peak” after 2023 because every model could now rewrite, shorten, and change tone. The numbers say something else. In May 2025 Grammarly said it had closed a $1 billion growth financing from General Catalyst and was at more than $700 million in annualized revenue, with 40 million users and over 50,000 business customers. That is not a dying product, that is a large, cash-generating distribution engine. Links:

What did change after GPT-4 and Claude 3 was where Grammarly’s value sat. Grammar- and tone-check moved toward free; the value moved toward being present everywhere (in Docs, Slack, Outlook, Chrome), managing style at org level, adding AI agents for research, originality, even authorship. That is why in January 2025 Grammarly bought Coda and brought in Shishir Mehrotra to run the combined company — to stop being a single-surface tool and become a suite. By mid-2025 they were already talking in public about “apps and agents, not just writing.” Source: Grammarly x Coda deal, Dec 2024 / Jan 2025, TechCrunch on the pivot

A company at $700m+ ARR with fresh, non-dilutive money and a CEO who has already built a productivity platform once is perfectly capable of buying a $30–40m ARR premium app.

Superhuman was hot — but not unbuyable

Superhuman did get a second wind from LLMs. Once AI replies, follow-up nudges and “write in my voice” landed, the product finally looked less like “faster Gmail” and more like “email plus an assistant.” 2025 reviews all say the same thing: $25–30 per user per month, clear premium positioning, strong differentiation on speed, and very sticky among founders, sales and exec teams. See:

But Superhuman’s 2021 $825m valuation was set in a zero-rate market. Reuters reported at the time of the acquisition that the company was doing “about $35 million annually” in 2025. That is good for a premium email app; it is not good enough to stay independent forever if the giants (Google, Microsoft) are bundling AI mail, and if smaller rivals like Shortwave, Hey and even Zoom Mail are adding similar AI features. At the same time, Grammarly was already rewriting 50 million emails weekly across 20 providers, just not in its own email client. Buying Superhuman gave it a native surface where those AI capabilities could be shown without fighting Gmail’s or Outlook’s UX. Source: Reuters on the acquisition, July 2025

So you had a buyer with cash, distribution and an email gap; and a target with product love, AI momentum and an addressable but not gigantic business. That is a classic fit.

Why a “threatened” company buys the “rising” one

The paradox here is just that we misunderstood who was threatened.

Generative AI threatened Grammarly’s feature moat — spellcheck, rewrite, tone — but not its distribution moat. Grammarly is still the extension that’s allowed in companies, the one students already have, the one HR doesn’t ban. Its AI agents get to ride that distribution. That kind of trust is extremely expensive to rebuild.

Superhuman, conversely, had an expanding feature moat thanks to LLMs — drafts, follow-ups, “respond like me” — but very narrow distribution: Gmail and Outlook power users willing to pay $30 a month. AI actually made the product better but didn’t make the market 10x larger. Grammarly could widen Superhuman’s distribution overnight by stuffing it into the same channel it already uses to sell to 50,000 orgs.

In other words: Grammarly lost some uniqueness but kept the pipes; Superhuman kept uniqueness but lacked the pipes. That is why the money flowed in the “wrong” direction.

Email is where their AI actually runs

There is also a product-vision reason. Grammarly has been telling investors since 2024 that the real battle is for “the surfaces where people actually write.” They already owned browser and docs. They didn’t own email. Yet email is where an AI assistant can do the most credible, measurable work: write replies, summarise threads, schedule meetings. That is what the July blog post said explicitly: “email is a critical communication surface in our vision of an agentic future” (Grammarly blog, June 30, 2025).

By October, when they rebranded the whole company as Superhuman and bundled Grammarly, Superhuman Mail, Coda and the new Superhuman Go assistant, they had a clean story: one subscription, all writing and comms surfaces, AI agents in each. Sources: The Verge, Oct 29, 2025, Engadget, Windows Central’s critique

That’s the part people missed when they said “but ChatGPT killed Grammarly”: OpenAI, Anthropic and Google don’t yet own the inboxes and docs and LMSes and HR portals where enterprise writing actually happens. Grammarly still does. Superhuman just gave it the prettiest, most opinionated inbox to drop its agents into.

So… was Superhuman actually in better shape?

It was in good shape, not unassailable shape. LLMs helped Superhuman on the user side, but they also made it easier for Gmail, Outlook, Zoho and even Notion to offer “good enough” AI replies. Superhuman’s answer was to go deeper on workflow: follow-up detection, VIP triage, “never drop the ball” nudges. That’s smart, but it is still a narrow market. Meanwhile, Grammarly’s answer to AI was to go horizontal — docs, email, web, now an agent store. Horizontal plus a billion dollars in fresh credit beats narrow plus love.

Seen in that light, the acquisition is not strange at all. It is the ecosystem rearranging itself so that the company with the biggest install base and the widest channels can keep showing up in front of users, even as the individual AI tricks become universally available.

And that is the hidden moral of mid-2020s AI: the players everyone assumes are “already dead” because GPT can do their core task often have the healthiest distribution. That distribution is exactly what the shiny, LLM-native apps end up selling to.