Superhuman’s new auto-draft feature almost makes me like AI replies

Superhuman, the premium email client, is launching a new version of its auto-draft feature that uses advanced AI from Anthropic and OpenAI to generate email replies that sound less robotic and more like the user. In beta testing, the feature has shown promising results: 40% of auto-generated drafts were sent within one day, and 60% of those were sent without any manual editing, according to co-founder Rahul Vohra.

How the new auto-draft feature works

The updated auto-draft feature identifies important emails and creates draft replies based on the user’s tone from previous conversations. It also generates two alternative responses. During a hands-on beta test, the feature successfully drafted replies agreeing to embargoes, confirming meeting times, and even declining requests for authored posts — all with minimal editing required. However, it sometimes generated overly positive responses to pitches or agreed to meetings at inconvenient times. Users can quickly select a different variation from the alternatives.

Learning from user behavior

The feature improves over time by learning from user edits. For example, after the beta tester rejected a midnight meeting suggestion, the system later generated a draft saying the timing didn’t work when a similar time was proposed. Users can personalize the AI further by adding details about themselves and their role in Settings > Personalization, and by uploading files or links for additional context.

Under the hood: Models from Anthropic and OpenAI

Vohra explained that earlier features like Instant replies relied on older models such as GPT-3.5, which had smaller context windows and produced less intelligent outputs. The new implementation uses a mixture of frontier models from both Anthropic and OpenAI. “The actual writing is done by frontier models from both Anthropic and OpenAI. So we’re applying the maximum amount of intelligence and context to this that we possibly can to make the feature work,” Vohra said.

Broader context: Superhuman Go

Last year, Grammarly acquired Superhuman and rebranded the company as Superhuman. The company is now building an assistant called Superhuman Go that spans across platforms while carrying context from one app to another. This auto-draft feature is a step toward that broader vision.

Why this matters for users

For professionals who receive thousands of emails monthly, AI-assisted drafting can save significant time without requiring full delegation of inbox management. The feature is designed to help users respond to more people when they don’t need to type long messages, while still maintaining control over tone and content.

Conclusion

Superhuman’s new auto-draft feature represents a meaningful improvement in AI-generated email replies, using advanced models to produce more natural-sounding drafts that learn from user behavior. While not perfect, it offers a practical tool for managing high email volumes without sacrificing personal voice.

FAQs

Q1: How does Superhuman’s auto-draft feature differ from previous versions?
The new version uses frontier models from Anthropic and OpenAI instead of older models like GPT-3.5, resulting in more context-aware and less robotic replies.

Q2: Can the AI learn my writing style?
Yes. The feature learns from your edits and previous conversations to adapt its tone and suggestions over time.

Q3: Is the feature available now?
The feature is currently in beta testing. Superhuman has not announced a public release date.