From: redpointai
This article explores the market dynamics, product strategies, and investment considerations for AI-powered solutions, drawing insights from Rahul Vohra, CEO of Superhuman. Superhuman is an email tool known for its speed and integration of artificial intelligence features [00:00:23].
The Enduring Role of Email and AI’s Impact
Despite perennial claims of its demise, email persists as a fundamental communication tool due to the foundational nature of email addresses as unique identifiers owned by companies [00:01:44]. The future of email is expected to be significantly shaped by two major themes: AI and collaboration [00:02:49].
Rahul Vohra believes we are heading towards an agentic AI future where AI agents will operate autonomously within email, handling tasks and interacting with other systems on behalf of users [00:03:02].
Superhuman’s AI Product Strategy
Superhuman’s approach to integrating AI has followed a phased roadmap:
- On-Demand AI Features (Phase 1): These features require users to remember to activate them. They are generally easier and cheaper to build and run, serving as an initial test for technology understanding, effectiveness, and user adoption [00:06:26].
- Write with AI: Launched in June or July of the previous year, this feature transforms a few words into a fully written email, matching the user’s existing voice and tone [00:04:19]. It also allows for shortening, lengthening, improving, and fixing writing [00:04:32].
- Always-On AI Features (Phase 2): These features operate continuously, working in the background for the user. They are more ambitious, difficult, and expensive to build and run due to operating at massive scale [00:06:56].
- Auto Summarize: Launched in November, this feature provides a one-line summary above a conversation, which updates instantly with new emails. Users can expand it for a bullet-point summary [00:04:41]. This changes user behavior, as often only the summary is needed, saving time [00:05:14].
- Instant Reply: Released recently, this feature drafts replies for every email, sometimes eliminating the need for user edits [00:05:30]. Users of instant reply write emails twice as fast, demonstrating significant business impact [00:06:05]. This feature also provides inspiration for replying to emails that might otherwise be ignored [00:09:11].
- Agentic AI Future (Phase 3): This phase builds upon the foundation of always-on features, envisioning a future where multiple AI agents, including an email agent, are mostly autonomous. These agents would be able to understand goals, break them into tasks, handle ambiguity, and interact with other agents or systems like CRMs [00:07:50]. This ultimately aims to free humans for more creative, strategic, and impactful work [00:08:32].
Operational Model for AI Development
Superhuman adopted a “Theta mode” operating model specifically for AI development, which involves an embedded executive doing individual contributor work daily to ensure the existential project’s success [00:14:12]. This allows for extremely rapid movement [00:14:12].
Key aspects of their development process:
- Iterative Design: Early on, there was no clear standard for AI user interfaces [00:14:52]. Superhuman experimented with different designs, such as whether replies should be visible before typing, eventually realizing that visibility provides inspiration to reply [00:19:11].
- Seamless Integration: A core principle is “when you want it and out of the way when you don’t” [00:19:38]. For instant reply, this meant short, snappy responses to match the user’s fast workflow [00:21:14].
- Interaction Design: The usability of a feature often comes down to minute details, like switching from arrow keys to
TabandEnterfor selection, making it 10x faster and more usable [00:23:00]. - Prompt Engineering: Instead of fine-tuning, Superhuman prioritizes extensive prompt engineering, feeding the model with examples of the user’s past emails to match their voice and tone [00:25:50]. They use Brain Trust for robust regression testing and evaluation frameworks [00:26:55].
- LLMs are used to self-assess replies for correctness (topic, person, length) [00:28:34].
- Model Selection: Open AI models (specifically GPT-3.5 Turbo for peak throughput) were chosen for their quality lead (6-9 months ahead), commitment to lowering costs, and collaborative approach with startups [00:31:04]. Throughput and latency are often more critical than intelligence for bursty, real-time features [00:32:45].
Pricing Strategy for AI Features
Superhuman has maintained its $30/month retail price for eight years, effectively shifting budget from initial concierge onboarding to powering AI features [00:34:12]. While considering future pricing adjustments for various product lines (e.g., Superhuman for Sales, Enterprise, Platinum Edition, or AI features), a definitive strategy is still in development [00:38:38].
Market Outlook for AI Startups
Rahul Vohra’s investment thesis for AI companies:
- Bearish on “Denovo” AI Apps: Startups building entirely new applications and integrating AI from scratch face the monumental challenge of essentially building two companies at once, making it very difficult without massive funding [00:41:36].
- Bullish on Distribution Advantage: Success favors companies with existing distribution advantages, such as strong contacts, viral loops, or a proven track record of user acquisition and retention [00:42:15]. AI might serve as a hook but is secondary to distribution [00:42:19].
Taking on Incumbents
While incumbents may seem to have an advantage in AI due to existing data and user bases, Vohra argues that startups still have significant opportunities:
- One-Size-Fits-All Solutions: Large incumbents like Microsoft (Outlook) and Google (Gmail) serve billions of users, forcing them to offer generic solutions [00:48:34]. This creates an opening for startups to target underserved, economically powerful market segments with superior, specialized products [00:48:58].
- Product Speed: Incumbents struggle with product speed due to legacy architectures. Superhuman, built entirely in JavaScript, achieved “instantaneous response” and “sub-100 millisecond user interactions,” a difficult feat for older client-server applications to replicate [00:50:00].
- Design and Conway’s Law: Incumbent product design often reflects organizational structure rather than optimal user experience [00:50:18]. This leads to disjointed experiences and features that prioritize internal team structures over user logic [00:51:07]. Startups can build products with a coherent, user-centric design from the ground up [00:51:07].
Market Sizing and Verticalization
Vohra advocates for starting with a specific user archetype and then expanding to multiple personas, rather than attempting to go broad from the outset [00:52:09]. For Superhuman, this meant starting with founders and VCs, then expanding to leaders, managers, and “outbound professionals” (salespeople, recruiters, consultants) [00:52:37].
The email market is enormous, with a billion professionals spending approximately 3 hours daily on email, totaling trillions of hours annually [00:54:33]. This scale allows for significant businesses to be built even by serving a small fraction of the market [00:55:01].
The Future of Work and AI Agents
AI’s Impact on Jobs (Overhyped and Underhyped)
Rahul Vohra views the idea of AI “coming for our jobs” as both overhyped and underhyped [00:55:47].
- Underhyped: The rapid displacement of certain entry-level roles (e.g., customer support, sales) is not widely understood outside of tech [00:56:06]. If AI can perform a task 80% as well at 10% of the cost, companies will adopt it en masse, leading to a better consumer experience than current robotic customer service [00:56:36].
- Overhyped: The broader impact of AI extends beyond these roles to all jobs, including those of CEOs [00:57:08]. Vohra believes AI will eventually perform most of his CEO duties better than he can [00:57:08]. This societal shift, where what we perceive as “work” today may be seen as “leisure” in the future, is happening much faster than historical shifts [00:57:50]. Humans are hardwired to work and engage in status games, meaning “leisure” activities will still feel like purposeful work [00:58:17].
Future of Work Agents
The agentic AI future will feature multiple, mostly autonomous agents. The “trillion-dollar question” is how this ecosystem of AI agents will come together and interact [00:44:03].
- Orchestration: A central service (like a future Chat GPT) may federate AI agents, allowing them to communicate and authenticate with each other (e.g., an executive AI talking to a Rippling AI to get health insurance information) [00:44:05]. This process could be instantaneous, saving days of human interaction [00:44:20].
- Product Placement: Products like Superhuman, which users spend 3+ hours a day on (similar to chat/messaging apps), are well-positioned as a “pane of glass” for interacting with these agents [00:45:33].
- Domain-Specific Agents: Each company is likely to build an agent relevant to its domain (e.g., an email agent that triages, drafts, schedules, and sends emails) [00:46:02]. These agents would then interact with each other within a broader ecosystem [00:46:04].
- Beyond Current Tools: Existing tools like Slack, despite their ubiquity, lack fundamental features that make email effective for managing conversations (e.g., ordered by last reply, subject lines, “done” status, snoozing, assigning threads) [01:00:00]. The future of workplace experience will combine the best of both worlds, offering choice of modality and organization for thoughtful asynchronous discussion [01:01:50].
- Personal Agents: A future of work agent could understand personal priorities, calendar availability, and preferences to organize daily tasks and reduce anxiety [01:03:22].
Vohra’s personal mission for Superhuman is to help professionals feel happier, more productive, and closer to achieving their potential [01:04:52].