Superhuman, the premium email client now owned by Grammarly, has launched a new version of its AI auto-draft feature that its creators say produces replies far less robotic than earlier attempts. Early beta testers report that the tool often generates responses requiring little to no manual editing, marking a significant step forward in AI-assisted email management.
The feature identifies important incoming messages and drafts replies based on the user’s tone from past conversations. It also provides two alternative drafts for each email. In testing, the system correctly handled tasks such as agreeing to embargoes, confirming meeting times, and politely declining requests for authored articles. However, the tool is not flawless — it initially generated overly positive responses to pitches and accepted meeting times after midnight. The system learns from corrections, so after the midnight meeting mishap it began flagging similar timings as unsuitable.
Superhuman co-founder Rahul Vohra explained that earlier features like Instant Replies relied on older models such as GPT-3.5, which had limited context windows and produced stilted language. The new version 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.
For UK businesses and professionals, this technology could reduce the time spent on email correspondence, particularly for those receiving hundreds of messages daily. However, concerns remain around data privacy and regulatory compliance. The UK Information Commissioner’s Office (ICO) has been increasingly active in scrutinising AI tools that process personal data, and the EU AI Act imposes strict requirements on high-risk AI systems. Companies using such features will need to ensure they have lawful bases for processing email content and that users retain meaningful control over automated outputs.
Dr. Elena Rossi, a digital ethics researcher at the University of Cambridge, noted that while the technology is impressive, it raises questions about authenticity in professional communication. “If a reply is generated by an AI, is it truly your voice? And what happens when the system makes a mistake that has legal or contractual implications?” she said. The feature allows users to personalise their profile with role details and reference files, but the ultimate responsibility for sent messages remains with the user.
Superhuman is now building Superhuman Go, a cross-platform assistant that aims to carry context between different applications. The company hopes this will further streamline workflows for power users, though it remains to be seen how regulators will treat such deeply integrated AI tools.