Superhuman's AI can now write your emails, and 60% of users aren't touching them
Curated by the Inblix editorial team
The promise of AI tackling your inbox has always felt more aspirational than practical—until maybe now. Superhuman is rolling out a new auto-draft feature that actually seems to understand which emails need a reply and how you’d personally write it, marking a significant shift from the robotic, over-eager responses that plagued earlier attempts.
I’ve been testing the beta, and the difference is real. The system generated drafts agreeing to embargoes on a pitch and confirming meeting times with minimal editing required. It even correctly declined requests for authored TechCrunch posts. Rahul Vohra, Superhuman’s co-founder, shared that during testing, 40% of auto-generated drafts were sent within 24 hours. More tellingly, 60% of those were fired off with zero manual editing. That’s not a trivial stat—it suggests genuine utility, not just a novelty.
The feature works by pulling your tone from past conversations and offering three response variations per email. It’s far from flawless, though. By default, it often defaulted to an enthusiastic “yes” for pitches or cheerfully agreed to meetings scheduled after midnight. But here’s where the learning kicks in. After I corrected the midnight meeting issue once, the system adapted and later auto-drafted a response saying the timing wouldn’t work when a similar request came through.
This isn’t powered by a single model. Vohra explained the company is now using a mix of frontier models from both Anthropic and OpenAI for the actual writing, applying what he called “the maximum amount of intelligence and context” they can. This is a notable architectural pivot from earlier features like Instant Replies, which relied on less capable models like GPT-3.5. The broader play here ties into Superhuman’s evolution following its acquisition by Grammarly and subsequent rebranding. The company is now building an assistant called Superhuman Go, designed to carry context across different apps, hinting at ambitions well beyond email.
💡 Key Takeaways
- Superhuman's new auto-draft feature sends 40% of its drafts within a day, and 60% of those require no manual edits, signaling a leap in practical AI email utility.
- The system adapts to user corrections—rejecting a midnight meeting once taught it to auto-decline similar requests later, showing genuine behavioral learning.
- Superhuman is architecturally shifting from single-model reliance to a mixture of frontier models from both Anthropic and OpenAI for maximum contextual intelligence.
- The feature is part of a larger Grammarly-backed vision for Superhuman Go, an assistant designed to carry user context across multiple applications.
Keep reading: See related articles below for more coverage on this topic.
Get smarter about AI
The sharpest AI news, curated daily. Delivered free to your inbox.