In June, Superhuman acquired GPTZero, a New York-based startup best known for creating one of the most popular tools for getting AI into student writing. Its young founders and team have joined the recently created “authenticity” group of Superhuman, where they are working to integrate the power of discovery into all the company's products, especially Superhuman Go, an AI assistant designed to work in everything the user does online.
The deal marks the latest step in a rapid turnaround led by CEO Shishir Mehrotra, who took over in early 2025 after the company, then called Grammarly, acquired its manufacturing startup, Coda. That summer, Grammarly bought Superhuman Mail, a powerful email tool built around the concept of “zero inbox.” In October 2025, Mehrotra renamed the combined company Superhuman, indicating a desire to build a suite of AI tools beyond a writing assistant.
Before founding Coda in 2014, Mehrotra built and launched products for YouTube and Microsoft. He repeats the standard Big Tech playbook—Google became Alphabet in 2015; Facebook became Meta in 2021—when a well-known product evolves into something more scalable.
“There have been enough instances of that being done in a way that preserves the basic type,” Mehrotra told the Observer at the time.
Grammarly, still Superhuman's most popular product, is used by 40 million people every day. Students rely on it to refine essays; editors use it to catch typos and enforce stylistic consistency—sometimes in full internal rulebooks that can run hundreds of pages.
Applying AI to writing—arguably one of the most human forms of work—has been controversial from the beginning: How much use of AI is too much? Is AI better used to produce the first draft or the final polish? And increasingly, can anyone reliably explain the difference?
Mehrotra had to navigate those questions publicly. Earlier this year, the company faced backlash over its “Expert Reviews” feature, which generated suggestions in the style of well-known authors and led to a class-action lawsuit over the license.
“There's a clear line, depending on the job, between things you should use AI for and things you shouldn't,” he told the Observer in an interview earlier this month. “In the case of a writer, it may be an excellent tool for researching or understanding a subject. But if it sneaks into his writing and takes away from the person making the report, it reflects badly on both the writer and the book.”
Educators and programmers have long complained that AI detection tools are unreliable, while many AI researchers argue that the problem is intractable. In any case, Mehrotra is already looking at the bigger picture. “Think of Grammarly as a high-quality English teacher sitting on your shoulder all the time, helping you everywhere you work. But in reality, it's a poor use of the infrastructure we've built,” he said. “The core infrastructure is the ability to bring embedded AI to your workplace.”
That vision underpins the company's rebranding and, not long after, its launch of Superhuman Go, which extends Grammarly's real-time help into a general-purpose system for agents.
“We allow anyone to build an agent that works like Grammarly, but does things that go beyond grammar,” explained the CEO. “For a salesperson, that might mean having a sales coach next to you, warning you that you're about to recommend the wrong product as you email a customer. For a support agent, it might mean a support coach reminding the customer that they had a great communication yesterday and suggesting you talk about it.”
Searching for human qualities in a world full of AI
Most of Superhuman's departments—from engineering to sales to customer support—use nearly every AI tool on the market, Mehrotra said. But when the information reaches the level of the CEO, he faces his own test in drawing the line for the use of AI.
“I often tell people: use AI to help you form your ideas, but not to write your summaries and reports,” he said. “When people send me something that clearly sounds like it was written by AI, I usually tell them, 'I'll just get informed.' I'd rather see your unique, raw understanding and start there, than read the output of a polished AI. “
The spread of AI has also changed the recruitment process. Tech leaders like Bill Gates predicted early in the AI boom that putting AI at work would be like having a personal digital assistant at all times. And that has changed what managers look at for job seekers, even for entry-level positions.
“AI has added another bar to almost every role; we're turning everyone into an early manager of their knowledge. You'll have a team working under you from day one, which is completely different from how it worked in the pre-AI world,” Mehrotra said. “All of a sudden, we have to look at very similar skills [to managers] even for entry-level candidates. Your ability to work with AI to produce the right result is remarkably similar to your ability to work with a team of human workers. “
All that remains is to search for unique human qualities. Mehrotra and his Coda co-founder, Matt Hudson, have pioneered the idea of using the discipline of the “eigenquestion” in decision-making—a loan from mathematics that reorganizes prioritization.
“We often find ourselves in situations where we have ten questions to answer and we have to prioritize them,” he explained. But when you use the eigenquestion test, you look at that list and ask: 'If I answer this one question, how many other questions on the list do you answer?' It would be the sixth question on the list of ten.”
This way of thinking is very important in making strategic decisions. As AI becomes increasingly capable of generating rational responses to everything, a scarce resource becomes deciding which problems to solve and which outcomes to trust.
“When I think about the misuse of AI, I think that people often try to insert human understanding with AI. When they do that, they end up, at best, embarrassing themselves, and at worst, they actually produce the wrong results,” Mehrotra said. “One of the things that is incredibly human is to judge from very little information. Ultimately, knowing what problem to solve also seems, at least for now, only human.”
