Since the explosion of large-scale language models (LLMs), companies are trying to solve the problem of overflowing inboxes by using AI to categorize emails and draft replies that look like you. Email client Superhuman releases a new version of its auto-drafts feature that identifies important emails and drafts less mechanical replies.
Superhuman has tried this in the past with features like instant replies and follow-up auto-drafts. However, many of the emails sounded like enthusiastic AI salespeople, so I didn’t use them much. The new version’s auto-draft feature feels different. Over the past few days, after accessing the beta version, I’ve sent out emails with few or no edits to the several drafts that were generated.
The app understands which emails require a reply and drafts replies based on the tone of previous conversations. It also generates two other variations that you can send instead.
In my experience using this feature, I’ve seen drafts that I’ve agreed to block from pitches to get more information, and meeting timings that I can send out with minimal editing. The feature also generated replies to emails asking me to submit my work to TechCrunch that I was not responsible for the work. (TechCrunch does not accept writing submissions.)
However, this feature is far from perfect. By default, they often responded positively to suggestions or agreed to meetings after midnight. Thankfully, I was able to quickly select and send another response from the other variations.
This feature learns from your usage and improves its response. For example, after a midnight meeting fiasco, when someone suggested a similar time, the feature generated a draft saying that timing didn’t work for me.
I receive thousands of emails each month, some of which are made possible by AI that makes first drafts easier for others, such as communications professionals and PR professionals. I’m not confident that I can hand over the reins to AI to completely handle my inbox, but this feature can help me reply to more people when I don’t have to type long messages.
Users can personalize their emails by going to Settings > Personalization and adding details about themselves and their role, as well as files and links for more context.
Rahul Vohra, co-founder of Superhuman, said that during the testing phase, 40% of auto-generated drafts were sent within a day, and 60% of them were sent without manual editing.
Vora said early features like instant replies were built from older models such as GPT-3.5, which were less intelligent and had smaller context windows. In this new implementation, the company is using a series of models.
“Currently, we are using a variety of models to achieve this functionality. The actual writing is done by frontier models from both Anthropic and OpenAI, so we are applying the maximum possible intelligence and context to this to make this functionality work,” Vohra said.
Last year, Grammarly acquired Superhuman and rebranded the company as Superhuman. Currently, the company is building an assistant called Superhuman Go that spans multiple platforms, handing over context from one app to another.
If you buy through links in our articles, we may earn a small commission. This does not affect editorial independence.
