Overview of 20VC: The $100 Billion AI Assistant Race
This episode features Harry Stebbings and JD, founder of Town, an AI assistant for email and calendar that automates work tasks for mainstream users and teams. The conversation zooms out from Town specifically to the broader AI assistant market: who wins, what the real moats are, how consumer and enterprise may converge, why model costs matter, and whether frontier providers like OpenAI, Anthropic, Google, Apple, Meta, and xAI will squeeze startups out. JD argues the market is moving extremely fast, that product-market fit matters more than theoretical moats right now, and that the eventual winners will likely be the products that become the user’s trusted “entry point” into digital work and life.
Key Takeaways
Town’s core thesis
- Town is an AI assistant embedded in email and calendar.
- It watches how you already work, then recommends and executes automations in the background.
- JD believes the key insight is simple but powerful: if you connect email + calendar, the product knows enough to be useful immediately.
The AI assistant market is still early
- JD thinks most products in the category still do not have deep mainstream product-market fit.
- Current tools may be impressive, but many are still power-user products, not mass-market utilities.
- He sees the market as a blue-ocean opportunity, but only for a small number of winners.
The biggest change: speed
- The market is now so fast that:
- startups can build at machine speed,
- but they can only learn at human speed.
- That means copycats can catch up quickly, shrinking the window for learning-based advantage.
Trust will shift from humans to agents
- JD predicts that in five years, users will trust their agent to:
- decide what data to share,
- maintain boundaries between personal, work, and sensitive information,
- and mediate information flow between coworkers, friends, and systems.
- He thinks this sounds risky today, but will become normal as models get better at respecting privacy rules.
Town’s Product Strategy and Go-to-Market
How Town works
- Town is designed around workflows already happening in email and calendar.
- The product identifies repetitive tasks and suggests automations.
- It aims for low time-to-value: users should get something useful quickly, with little setup.
Why it converts
- JD says a major reason Town converts well is that it has a hard onboarding requirement:
- users must connect email and calendar.
- That creates enough context for the product to deliver value fast.
- He says this is part of why Town has a strong trial-to-paid conversion rate.
Pricing and monetization
- Town’s plans are roughly:
- $15
- $49
- $99
- $199
- JD says the lowest tier is the most subsidized, while the $99 plan is likely the most profitable.
- He prefers a paid, value-aligned model over ad-supported AI, arguing ads create bad incentives in assistant products.
Customer success definition
- JD defines success very simply:
- if the customer keeps paying every month, Town is doing its job.
- He rejects token-maximization as a metric, because it can encourage wasteful use that damages trust.
Competition, Moats, and the “Winner-Take-Most” Race
JD’s view on moats
- He argues that talking about moats is almost a luxury at this stage.
- In his view, the market is too early and too fast for abstract moat theory to matter more than:
- product-market fit,
- user trust,
- and distribution.
What might actually become defensible
JD sees three possible sources of defensibility:
-
Agent-level network effects
- Town’s “agent-to-agent” feature lets your assistant query coworkers’ assistants.
- That creates a multiplayer dynamic that gets stronger as more people on the team use it.
-
Context and personalization
- The more a product understands a user’s tools, habits, and workflows, the harder it is to switch.
-
Brand and opinionated UX
- Town gives users a named, branded assistant (“a townie”).
- JD thinks that emotional relationship may matter more than many people expect.
The real threats
- JD is most concerned about:
- Meta/WhatsApp for personal use cases, because they already own distribution.
- Apple and Google because they own the device layer.
- OpenAI and Anthropic because they can move quickly and already have model/capability advantages.
- He also sees xAI/GrokBot as a more direct competitive threat than some consumer-first tools because of overlapping use cases.
Why device owners matter
- JD believes the long-term “entry point” to digital life may become the AI itself, not apps or websites.
- If that happens, whoever owns:
- the device,
- the distribution surface,
- or the assistant interface may have an enormous advantage.
Model Strategy, Cost, and Economics
Model routing is invisible to users
- Town routes tasks across models based on cost and performance.
- JD says most users do not care which model is used; they care that the result is good.
UX consistency matters
- For voice and assistant personality, consistency is crucial.
- If the model suddenly sounds different or behaves differently, users notice immediately.
- For pure reasoning or back-end work, model switching is easier.
Frontier costs are the biggest long-term concern
- JD says Town’s biggest economic risk is the share of tasks that remain close to the frontier, where pricing power is limited.
- For many routine tasks, cheaper/open models will eventually work well enough.
- But if a meaningful chunk of workloads must stay on expensive frontier models, margins get squeezed.
Current spend
- JD says Town’s run rate is around $75K per engineer in AI tooling.
- Their internal stack reportedly uses:
- Codex
- Claude
- Cursor
- Devon
- He believes AI is already making engineers more productive enough that hiring economics are shifting.
Enterprise, Consumer, and the Future of Assistants
One user, multiple entry points
- JD thinks most people will want one primary AI entry point, not 50 separate agent experiences.
- But due to privacy and workplace boundaries, there may still be:
- a personal assistant layer,
- a work assistant layer,
- and perhaps a device-level interface.
Consumer and enterprise may converge
- He doesn’t think consumer and enterprise are the same product today.
- But he believes the underlying agent layer may converge over time, with different permissions and data boundaries underneath.
Workflows will become more agentic
- JD expects assistants to handle:
- scheduling,
- email triage,
- pre-briefs,
- cross-team lookups,
- research,
- and eventually more complex multi-step tasks.
- He also thinks AI will increasingly help users do more revenue-generating work, not just save time.
Product, Team, and Operating Lessons
What Town learned from expansion
- A “tinkerer” on the team can accelerate adoption by creating:
- team routines,
- team skills,
- and integrations that spread across coworkers.
- Some of Town’s best growth comes from power users enabling others.
Underserved functions matter
- JD highlighted strong adoption in less obvious functions, including:
- executive assistants,
- chiefs of staff,
- HR,
- finance,
- recruiting.
- These users often still live in email-heavy workflows and are hungry for automation.
Time to wow is critical
- Town’s low time-to-value is one of its biggest advantages.
- JD says users often discover value almost immediately, then keep exploring.
Hiring philosophy
- Town has a very trust-based hiring style:
- if a trusted employee has worked closely with someone and strongly recommends them, they may skip the traditional interview gauntlet.
- JD’s logic: a trusted high performer is usually best placed to judge another high performer.
Notable Quotes and Ideas
On the pace of the market
“You can build now at the speed of machines, but you can only learn at the speed of humans.”
On the future of data sharing
“I think you’ll trust your agent to decide what data to share with other people without you intervening in five years.”
On the product race
“The speed of the market is insane… I’ve never seen anything like it.”
On business success
“If they keep paying me, I’ve done my job.”
Final Outlook
JD is broadly optimistic about AI’s impact on work and society. He believes the long-term outcome will be:
- more productivity,
- more revenue per worker,
- more personal leverage,
- and less day-to-day toil.
At the same time, he’s clear-eyed about the risk:
- frontier providers are powerful,
- distribution giants are dangerous,
- and the economics of AI assistants may get squeezed if too much work stays near the frontier.
His core conviction is that there is still a giant opportunity in AI assistants—but only for companies that can combine:
- strong product-market fit,
- fast iteration,
- trust,
- and a real path to durable value creation.