Superhuman's AI email assistant is getting good enough to actually use
A new auto-draft feature uses frontier AI models to write replies that match your tone. In testing, 60% of generated drafts were sent without any editing at all.
Key points
- Superhuman launched a beta auto-draft feature that writes email replies in your personal tone.
- During testing, 40% of AI-generated drafts were sent within one day, and 60% of those went out without manual edits, per co-founder Rahul Vohra.
- The feature draws on frontier models from Anthropic and OpenAI, replacing older GPT-3.5-based reply tools that produced stiffer text.
- Users can personalise how the AI writes by visiting Settings, then Personalization, and adding details about themselves and their role.
For anyone buried in email, the promise has always been the same: let AI handle the easy replies. Superhuman is now making a serious attempt to deliver.
The beta, first reported by TechCrunch AI, reads your inbox, identifies which messages need a reply, then drafts one aimed at sounding like you rather than a form letter. Two alternative versions are offered so you can pick the tone that fits. The feature learns from past conversations and adjusts based on which drafts you accept or ignore. In one beta example the system agreed to a post-midnight meeting; after the user sent a correction, it began declining similar requests on its own.
The step forward is the underlying technology. Earlier reply tools ran on GPT-3.5, a model with a narrower understanding of context. The new version uses the latest models from both Anthropic and OpenAI.
"The actual writing is done by frontier models from both Anthropic and OpenAI," Vohra said. "We're applying the maximum amount of intelligence and context to this that we possibly can."
The statistics behind that claim are modest but real. Forty percent of auto-generated drafts sent within a day is not a majority. Sixty percent of those going out unedited, though, suggests the writing is clearing a bar that earlier AI reply tools rarely reached. Short replies, meeting confirmations, standard declines: these are the emails that eat time without requiring much thought. Vohra's numbers say the feature handles a meaningful share of them without intervention.
This space is moving fast. When OpenAI launched ChatGPT Work in July, the pitch was similar: AI that absorbs context across your working tools and acts on it. Superhuman is making the same argument from inside your inbox.
Should ordinary email users care about this?
Not yet. The feature is in beta and not available broadly. When it does roll out, it lives inside Superhuman, which already requires a paid subscription. Grammarly acquired Superhuman last year and has kept the Superhuman brand.
For those who get access, the practical upside is real. The feature is not fully reliable: during early testing it generated overly positive responses and agreed to unsuitable meeting times. Checking the draft before sending remains essential.
Superhuman is also building a broader assistant called Superhuman Go, designed to carry context across multiple apps. That product is still in development.
The honest read here is that Vohra's numbers are encouraging without being decisive. A feature that handles 60% of accepted drafts without a single edit is worth watching. One that still books midnight meetings without prompting is not yet worth handing your inbox to.
