Interview
Town vs Instinct vs GrokBot | Why the AI Assistant Market Is Not a Bubble
- The product category is expected to be a top three priority for Google and Apple within the next 12 months, with the winning solution requiring a network effect at the agent level.
- Market competition is projected to accelerate, with competitors replicating features within two to four weeks, leading to a consolidation where only two or three startups will survive out of an initial 15.
- Apple's agent roadmap faces hurdles including limited cloud DNA, on-device privacy constraints, and slower local models, while the new Siri is predicted to be significantly less capable than Town or Grokbot for approximately nine months.
- Users are expected to trust agents to autonomously decide on data sharing for sensitive information like medical history or salary without human intervention within five years, driven by AI's lower error rates in tasks like email management.
- Privacy will dictate data silos, separating personal and work data, with Meta and WhatsApp emerging as major competitors in personal use cases due to distribution advantages.
- The industry is predicted to transition to a point where humans no longer read every line of code, as computers will build code with guardrails, while the percentage of tasks requiring frontier models remains the primary unknown economic variable.
- Compute prices are anticipated to halve every 9 to 12 months, enabling 20-30% product margins in 18 to 24 months, though scaling voice capabilities to 2-4 million users could negatively impact margins due to frontier model costs.
- Open weight models are expected to reduce costs by handling non-frontier tasks like labeling, while the $99 plan is forecast to be the most profitable segment compared to the worst unit economics found in the subsidized $15 plan.
- The $15 plan serves as a strategic entry point to transition users to the $49 plan, despite a 30% immediate churn rate when requiring email and calendar connections; the $99 plan balances revenue with controlled token spend.
- Business workflows offer larger revenue per user growth potential than the personal sphere, with free usage deemed unsustainable due to unpredictable costs and a lack of value validation.
- To reach $100 billion valuation, the Town agent requires approximately 10 million paying users, supported by a board member with CFO-like operational experience to manage scale and token economics.
- Engineering costs related to tooling, currently around 5% of total, will increase as better models replace human tasks, yet AI efficiency may drive hiring of incremental staff due to increased marginal profit per engineer.
- The market remains a "blue ocean" with high skepticism toward ad-backed models due to token costs, favoring paid-for assistance models to avoid polluting agent trajectories.
- European local competitors are expected to struggle unless they leverage specific distribution, GDPR, or regional moats, given the massive R&D spend required to compete with capability leaders.
- Defensibility strategies include "name and image" approaches similar to Snapchat and "time to wow" generation for single users, while pre-processing user mental models aims to improve work networking outcomes.
- Growth drivers in enterprises are identified as underserved functions like executive assistants, HR, and finance, while a "tinker" on a team can drive faster adoption through organizational routines.
- Long-term industry trends predict a shift to one or two entry points for digital interaction, increased generation of revenue per user via AI handling tasks, and the ability to do more with less toil.
- Future applications include AI accelerating drug discovery and physical construction, while self-driving capabilities may enable people to live further from cities with larger properties.
- Strategic investment focus remains on Base 10 or Modal, with the company currently in "growth investor land" before needing to prove detailed metrics for later funding stages.
- The "hard paywall" of connecting email and calendar is deemed necessary despite a 30% drop-off to unlock value, while parents and families show strong product-market fit for scheduling but lack the monetization willingness of mid-market firms.
- Integration of agents into X via Grokbot offers distribution benefits for power users but is not viewed as the primary path to winning the mainstream market.
- Competition between Town and Astro exists, but the market is large enough to accommodate multiple players if they differentiate, though "dark thoughts" about cannibalization by big providers persist.
- The ultimate business test remains the ability to generate real revenue from companies paying for products, with hiring processes increasingly relying on peer referrals from trusted team members.