readonrisk

a16z (Andreessen Horowitz) · Oct 19, 2026 · 62 min

Why a $1M Startup Wasn’t Worth Building in the Age of AI

Town CEO Jean Denis explains why his SMB tax startup was not venture-scale and how he pivoted to an AI assistant that works from email and calendar. He argues AI capabilities are converging across products, so the winners will be neutral, customizable apps that own user context, with agents increasingly replacing human-speed workflows. The conversation also covers why AI is diffusing unevenly, the threat from labs, and how agents may erode supply aggregators like Amazon.

Watch on YouTube ↗More from a16z (Andreessen Horowitz)

Where they stand

AI

Very bullish

AI intelligence is already sufficient and keeps improving quickly, with the main bottleneck being deployment and product experience, so adoption in business is a matter of when rather than if.

High conviction · years

Markets

Bearish

Supply aggregators like Amazon and thin app-layer software businesses face value erosion as agents automate human attention and abstract away apps.

Medium conviction · years

Key takeaways

  1. Town's SMB tax product worked technically but wasn't a venture-scale business post-AI: annual-only usage, little word of mouth, and incumbents able to automate tax for existing customers faster than a startup could take share. 4:20
  2. Town found product-market fit by building AI around email and calendar workflows, helped by models becoming more agentic around Opus 4.5. 11:28
  3. Denis sees AI product capabilities converging, with the app layer losing value and differentiation coming from per-user customization, learned from power users' routines. 29:29
  4. Town positions itself against labs on model neutrality, customer ownership of data and context (switching in a day), no training on customer data, and trusting agents to mediate internal requests. 37:09
  5. AI diffusion is held back by a lack of 'tinkerers' and text-box products; the intelligence is already sufficient and the problem is deployment, which requires showing value and doing the work for users. 47:02
  6. Cheaper tokens open up long-tail tasks (prep notes for every meeting, wholesaler price monitoring) that were never worth doing with humans, a Jevons-paradox dynamic for assistants. 23:28
  7. Agents may make supply aggregation like Amazon less valuable as human attention is automated away; Denis argues Amazon's best countermove is to have its own personal agent rather than block others. 57:58
  8. The labs' forward-deployed engineer approach looks good now but should get worse as models improve, per Denis, while automated custom software scales better. 32:49

Calls and positions

  • The AI market for helping businesses do more is bigger than coding and will move the needle at large AI labs within about 12 months.
    “in the current environment, it's not going to move the needle at those companies, but I think in 12 months it will”
    Very bullishPrediction: general business AI becomes material for labs (in 12 months)Jean Denis, Oct 19, 2026Watch at 43:46#OpenAI#Anthropic
  • The forward-deployed-engineer approach used by labs will get worse as models improve, while automation-based custom software scales better.
    “as the models get better, the FDE approach actually gets much worse”
    BearishPrediction: FDE model weakens as models improveJean Denis, Oct 19, 2026Watch at 32:49#OpenAI
  • The future involves far less human in the loop, with economic benefits to companies outstripping the risks.
    “I think the future looks like a lot less human in the loop than we see today”
    Very bullishPrediction: less human-in-the-loopJean Denis, Oct 19, 2026Watch at 41:33#AI agents
  • Amazon will lose to agents and its best countermove is to build its own personal agent rather than block others.
    “Amazon might lose a huge chunk of their advertising business trying to fight the future”
    BearishPrediction: Amazon loses advertising share unless it builds an agentJean Denis, Oct 19, 2026Watch at 57:58
  • Most people will take 6 to 18 months to experience the magic of personal AI assistants, as with the iPhone.
    “it will probably take 6, 12, 18 months for most people to experience this magic even in their personal life”
    Very bullishPrediction: mainstream personal assistant adoption (6 to 18 months)Jean Denis, Oct 19, 2026Watch at 18:32

Views expressed

  • The unit of value in software has shifted from the app to the computer, so value in the app layer will drop.
    “we're abstracting away the app layer totally and I think the value will drop in the app layer”
    BearishJean Denis, Oct 19, 2026Watch at 29:29
  • AI is not diffusing because the product experience is a text box and most people lack tinkerers, not because intelligence is lacking.
    “the product experience can't just be a text box”
    BullishJean Denis, Oct 19, 2026Watch at 47:02#OpenAI#Anthropic
  • Computer-use and browser-use agents now succeed about 90% of the time on tasks that succeeded roughly 30% of the time six months earlier.
    “in the last two and a half months the assistants can do things that were near impossible six months ago”
    Very bullishJean Denis, Oct 19, 2026Watch at 50:52#AI agents
  • Supply aggregation is much less valuable now because human attention can be automated away.
    “supply aggregation is a lot less valuable because human attention can be automated away”
    Very bearishAlex Rampell, Oct 19, 2026Watch at 1:01:14
  • Enterprise adoption of workplace AI will lag SMB and mid-market adoption because security, audit and cost-control requirements move at human speed.
    “As you move down market, right? As you move towards SMBs and mid-market firms, I think you have fewer of those requirements”
    BullishJean Denis, Oct 19, 2026Watch at 20:43
  • Cheaper tokens mean tasks like meeting prep notes become worthwhile for an LLM at about a penny, which humans never did.
    “But an LLM can do that for, I don't know, a penny, right?”
    Very bullishJean Denis, Oct 19, 2026Watch at 24:00

Topics covered

More from a16z (Andreessen Horowitz)

Summary generated by AI from the video's captionsand checked against it; quotes are verbatim. Processed Oct 25, 2026.

Read on Risk summarises what other people said in public videos and podcasts. Summaries are generated by AI and checked by code against the transcript, but can still be wrong; follow the source links before relying on anything. This is not investment advice and nothing here is a recommendation to buy or sell. Disclaimer ·How this works