Superhuman's founder-led sprint to build AI email
Superhuman bet that AI would remake email and ran the project as a founder-led tiger team rather than a normal product initiative. With no machine learning staff, it shipped on-demand features first, then always-on features that run constantly, and is now building toward an agent that can triage, schedule and write. The business case rests on responsiveness as part of the job and on selling to teams, which are stickier than individual subscriptions.
Takeaways
- AI became an existential priority for Superhuman. Vohra says it became clear in January or February that AI was an existential moment and everything would change. Superhuman made AI one of its three company priorities.
- Founder-led tiger team built AI without ML staff. Superhuman had no machine learning engineer and had never used large language models. Vohra ran the project himself with a designer and a few engineers, assigning no product manager, and was an individual contributor for six months.
- Email is collaboration, so Superhuman built multiplayer. Users can share a live view of an email with teammates, who see new messages even if they are not recipients, and comment in threads stored separately. Building this took well over a year after years of single-player work.
- Selling to teams beats individual subscriptions. Vohra says few software companies scale on individual subscriptions alone, and those that do must add tens or hundreds of millions of subscribers yearly because they retain only 50 to 60 percent. Teams are stickier and enable negative churn.
- Responsiveness is part of the job. Vohra says people who do email for a living, including founders, account executives, recruiters and realtors, treat responsiveness as part of the job. Superhuman measures time saved against Gmail and Outlook and through its product-market-fit survey, which produced the four-hours-a-week figure.
- Superhuman is building toward an agentic email future. Vohra imagines an agent that triages inboxes, schedules meetings, writes and sends emails, breaks problems into sub-problems, resolves ambiguity and interrogates internal systems. He expects this to arrive in some places over the next year or two, starting with Ask AI.
Superhuman's AI confidence ladder
- On-demand AI runs only when a user triggers a feature and is cheap to run. Superhuman's first AI feature worked this way.
- Auto summarize was the first always-on feature. It precomputes a one-line summary of a conversation that updates as emails arrive.
- Instant reply drafts a reply to every email.
- The team treated each step as a confidence ladder, taking on increasingly ambitious and expensive projects.