Superpowered, not superseded: The human rhythm of AI

Superhuman’s AI Revolution: Empowering Every Workflow with Proactive Intelligence

Superhuman, formerly Grammarly, is transforming into an AI-native productivity suite. CEO Shishir Mehrotra highlighted its impressive scale: 40 million users, $700 million revenue, and 100 billion LLM calls weekly. This rebrand reflects a broader ambition to embed AI into every user interaction, moving beyond just writing tools.

Mr. Mehrotra explained the “last mile problem” in AI, where users often switch applications for AI features. Superhuman’s core strength is embedding AI directly into over a million applications, including web browsers and desktop software. This seamless integration ensures AI assistance is always available within the user’s workflow, reducing friction.

The company launched Superhuman Go, a platform for numerous AI agents. It debuted with hundreds of agents for writing tasks and integrations with CRM, email, and support systems. The platform is open, allowing external developers and authors to publish specialized AI agents, expanding the ecosystem’s capabilities.

Superhuman Go’s core platform is initially free, with a premium tier planned. Individual agents can be free or paid, establishing an “agent store.” This model creates new business opportunities for developers to monetize specialized AI solutions, catering to diverse user needs and fostering innovation.

Mr. Mehrotra contrasted Superhuman’s “assistive” AI with “chat” (like ChatGPT) and “do” models. Superhuman’s AI is proactive, offering help before users even ask, powered by 100 billion weekly LLM calls. This makes it a leading AI player in individual user engagement, providing orders of magnitude more interaction.

The company emphasizes specialized agents over generic AI. While underlying technology is broad, agents are highly specific and context-aware. For instance, a sales agent connects to a user’s CRM, understanding their specific customers and products, offering personalized assistance rather than general advice. This deep integration ensures relevant support.

Superhuman’s long history with AI has shaped its optimization. While 3% of LLM calls use public models, 97% are handled by their own hosted LLM models, including open-source Llama. This internal focus prioritizes speed and cost-effectiveness, crucial for delivering real-time, embedded assistance efficiently.

Executive Editor Jim Edwards inquired about competition from general-purpose AI like ChatGPT. Mr. Mehrotra clarified that Grammarly often complements such tools, working in concert to enhance user experience by offering proactive, embedded assistance rather than requiring explicit prompts.

Mr. Mehrotra is enthusiastic about AI in education, envisioning “digital twins” of professors providing instant, continuous feedback to students, transforming the student-teacher dynamic. He views the AI landscape as a “bubble” that, despite potential failures, is crucial for driving infrastructure development and fostering future groundbreaking companies.