20VC: The $100 Billion AI Assistant Race: Town vs Instinct vs GrokBot | We Spend $75K Per Engineer on AI Tools | Why the AI Assistant Market Is Not a Bubble & AI Assistants Will Replace Every App on Your Phone with JD, Founder of Town

What It Covers

JD, founder of Town.com and former Plaid CTO, breaks down the AI assistant market with Harry Stebbings — covering competitive dynamics between Town, Instinct, and GrokBot, unit economics at $75K per engineer on AI tooling, model routing strategy, network effects at the agent level, and why the category will replace every app on your phone.

Key Questions Answered

• •Agent-level network effects as the real moat: Town's "agent-to-agent" feature lets one user's AI query a coworker's AI directly — bypassing human bottlenecks entirely. JD argues no one has cracked multiplayer AI yet, and whichever product does will be nearly impossible to displace. Once a full team operates on interconnected agents sharing context, switching costs become structural rather than superficial.
• •15% paid conversion via forced data connection: Town requires users to connect email and calendar before accessing the product — a hard gate that causes 30% immediate drop-off but drives over 15% of remaining users to pay. The mechanism works because upfront data access enables Town to suggest specific automations tailored to each user's actual workflows, creating immediate perceived value versus generic chat interfaces.
• •Frontier model dependency is the existential economic risk: JD identifies the core long-term threat: if 20–30% of workloads permanently require frontier models, AI assistant companies pay suppliers who are also direct competitors — OpenAI, Anthropic — at margins those suppliers control. He prices products today targeting 20–30% gross margins in 18 months, betting that non-frontier task costs halve every 9–12 months as open-weight models improve.
• •$75K per engineer annually on AI tooling is already ROI-positive: Town spends roughly $75K per engineer per year across Devin, Cursor, Codex, and Claude. JD frames this against fully-loaded Silicon Valley engineering costs including equity, concluding the compute spend equals roughly 1.5 engineers in value delivered. With revenue opportunities ("gold") everywhere, the constraint is execution speed — not whether AI tooling pays for itself.
• •Prefer 100M users at $20/month over 1M at $100/month: JD explicitly chooses volume over ARPU because business users generate compounding revenue — a recruiting firm paying $600/month gains one additional client worth $3,000/month, creating clear ROI that drives organic upsell. Personal use cases have natural consumption ceilings. Business use cases expand as AI delivers more measurable output, making NRR expansion the primary growth lever.
• •Apple's agent roadmap faces two structural disadvantages: Apple's on-device privacy commitment limits cloud data access, and agents perform better with more data — not less. Their non-cloud DNA compounds the problem. JD estimates Apple's new Siri will be roughly nine months behind Town and GrokBot in capability at launch, and argues Apple needs leadership with fundamentally different instincts to compete in the agent generation.