The Future of Email: Superhuman CTO on Your Inbox As the Real AI Agent (Not ChatGPT) — Loïc Houssier

Episode Summary

Executive Summary: Superhuman’s CTO explains how AI is being used to make email more proactive, faster, and more useful—via auto-labeling, summaries, drafting, deep search, and agent-like workflows—without adding latency. He argues the future of communication is conversational, memory-rich, and potentially voice-first, and that Superhuman’s advantage comes from quality, offline-first performance, and contextual understanding across email and adjacent tools.

Main Topics: AI as productivity acceleration in email (Priority: 5/5): Superhuman’s AI strategy is centered on practical productivity gains: classifying mail, summarizing threads, detecting unanswered emails, and pre-drafting responses so users can act faster with minimal friction. Agentic search, drafting, and workflow automation (Priority: 5/5): The team is building tool-based agents that can search deeply across inbox history, infer intent, handle follow-ups, and execute specific tasks like finding availability or drafting availability replies. Evaluation, quality, and model selection (Priority: 5/5): The CTO emphasizes rigorous evals across dimensions like deep search, date handling, and handoff behavior, and says model selection is driven first by quality, then by cost optimization. Offline-first architecture and data storage (Priority: 4/5): Superhuman keeps local email copies and search capabilities to ensure speed and offline use, while relying on Gmail/Outlook for source data and external providers for inference and embeddings. The future of communication: voice, conversational UI, and EA-like assistants (Priority: 5/5): He argues communication may shift from typed email to spoken or conversational experiences, with AI eventually acting like an executive assistant that knows preferences, context, and priorities. Team structure, engineering productivity, and AI adoption (Priority: 4/5): Superhuman is a small, senior team that saw meaningful engineering throughput gains from AI tooling, while also carefully managing quality, onboarding, and support implications. Broader product and market strategy (Priority: 3/5): The conversation covers Superhuman’s position versus OpenAI/Google, the importance of contextual awareness across apps, and the possibility of new product categories like browsers and knowledge graphs.

Key Arguments: AI should be embedded only where it measurably improves user productivity; “sparkles” without utility are not enough. Model quality matters more than cost initially; optimization comes after the user experience is working well. Deep inbox search requires iterative, paginated, semantic retrieval because relevant answers can be buried across years of mail. Agentic systems need many specialized tools rather than one monolithic agent. Evaluation must be multidimensional, covering handoff quality, deep search, date reasoning, and other use-case-specific behaviors. Superhuman’s offline-first, local-storage architecture is essential to its speed and user experience. The future of email may be conversational and voice-based, with AI acting as a proactive assistant rather than a reactive chatbot. A generic knowledge graph is hard because productivity data and entities are highly subjective across users and companies. AI increases engineering leverage and helps strong engineers ship more, but fundamentals still matter; weak engineers may be exposed faster. Superhuman’s long-term vision is to become an AI executive assistant for high-value users, not merely an email client.

Data Points: Team size: ~50 engineers - Superhuman’s overall engineering organization Paying users: ~100,000 - Approximate paying user base mentioned for Superhuman User email volume: 500–1,000 emails/day - Rahul, the CEO, was cited as receiving this volume Offline sync window: Last 30 days - Emails downloaded to device on install for offline search Retention horizon on device: Up to 2 years - Local email history is optimized over time on device Latency target: Under 100 ms - Every interaction is designed to feel instant PR throughput Q1: 4 PRs/engineer/week - Baseline engineering throughput before broader AI adoption PR throughput Q2: 5 PRs/engineer/week - Measured increase after AI/tooling adoption and process improvements PR throughput Q3: 6 PRs/engineer/week - Further throughput increase in subsequent quarter AI adoption in PRs: ~80% of engineers flag AI use - Engineers self-report AI usage on pull requests Positive AI impact among AI-flagged PRs: ~90% - Most self-reported AI-assisted PRs were viewed as productive Substack/newsletter load: 30–40 subscriptions - Example of how users use Ask AI to summarize newsletters Cost framing: $200/month willingness to pay - Some high-value users said they would pay more for the best model

Pivotal Quotes: "We as a human species, like we started to write because we didn't have like enough storage for the stories that we were telling to each other." — Loi Kusier: Opening reflection on why communication may move from writing to voice and AI-mediated formats "The main driver is how you can put AI in the product to accelerate the productivity of people." — Loi Kusier: Core statement of Superhuman’s AI philosophy "AI will do only one thing: it will separate faster the good engineers from the bad engineers." — Loi Kusier: Closing view on AI’s impact on software engineering and developer skill

Implications: Email and knowledge work are moving toward proactive, AI-assisted, conversational interfaces. Winners will combine strong UX, memory/context, offline speed, and high-quality evals—not just raw model access.