How Superhuman built and shipped AI features in four months

Executive overview

Most companies treat a major new technology wave as a project to plan. Superhuman treated it as existential and stripped away process to move fast. Rahul Vohra ran the AI initiative personally with a small tiger team, bypassing normal product management structures.

Founder fiat — making a decisive call without data, then owning the outcome — is what breaks decision paralysis at speed.

Building the AI feature ladder

• Started February 2023 with no ML engineers and no prior LLM experience
• Vohra personally led a small tiger team: himself, one designer, a handful of engineers
• Operated like an internal startup — no PM, no formal process, regular build sessions
• First shipped on-demand AI: write an email from bullet notes, matched to the user's own voice and tone
• On-demand features are cheap to run — only active when the user triggers them
• Success gave confidence to tackle always-on AI, which runs continuously in the background
• Always-on is harder to build and more expensive to operate
• First always-on feature: auto-summarize — a pre-computed one-line summary on every conversation, updating instantly as new emails arrive
• Second always-on feature: instant reply — every inbox email pre-populated with a draft reply on arrival

Why people pay for productivity software

• Target users do email for a living: founders, CEOs, AEs, recruiters, realtors, managers
• Responsiveness is a job requirement for these roles — slow replies block teams or damage reputations
• Superhuman measures productivity three ways: time saved, volume of emails handled, speed of response
• Users respond to 2.4x more emails in the same time compared to Gmail
• Sam Altman's YC observation: billion-dollar founders reply within minutes; failing founders take days or weeks
• Four hours saved per week is the headline metric, triangulated bottom-up and top-down via surveys

The case for selling to teams over individuals

• Single-player subscriptions are a weak foundation — very few software companies scale large on them alone
• Individual subscribers churn at 40–50% annually; only companies like Spotify and Netflix sustain it by adding tens of millions of new users each year
• Selling to teams is harder to close but produces far stickier retention
• Negative churn: seat expansion within a cohort outpaces seat losses, growing revenue per cohort over time
• Superhuman spent years making a product individuals loved before adding multiplayer features
• "I didn't want to build an email client your boss forces you to use"

Email as a collaboration tool

• Superhuman re-imagined email as multiplayer: any email can be shared as a live view with teammates
• No more forwarding, CC chains, or screenshots — team members see new arrivals in real time
• Threaded comments on emails are stored separately and never leak as email objects
• The rebuild took over a year and was a significant engineering lift
• Team accounts have scaled from SMBs to customers with several thousand seats

AI agents and competition

• Vohra's view on competition: focus on making customers happy, not on rivals — new email clients have come and gone repeatedly
• Superhuman's roadmap targets an AI agent that can triage, schedule, draft, and eventually send emails autonomously
• Agents will handle complex problems, break them into sub-problems, resolve ambiguity, and interface with APIs and other agents
• The agentic future is 1–2 years away in meaningful form; Superhuman is building toward it now