a16zWhy a $1M Startup Wasn’t Worth Building in the Age of AI
a16z
Tax startup rethink 0:00
After leaving Plaid, Jean Denise spent a few months with family and then started a B2B tax product. He saw tax as a place where AI could help because the work was text heavy and filled with structured documents like financials and lease files. The team proved it could automate the path from those inputs to a business tax return.
Why tax would not scale 4:02
The company found product fit, but not a path to venture scale. Tax could only be sold once a year, word of mouth was weak, and incumbents could use the same technology to reach their customers faster. By the end of the first season, the team decided the market would not let them reach the growth they wanted.
Moving to email and calendar 10:31
The founders then asked why no one had built a strong AI agent for email and calendar. Customer complaints pointed to meeting prep, meeting follow-up, simple replies, and daily readiness. As model abilities improved, the new product caught on fast. By December through April, it was growing daily and users kept pushing for more.
Product first for enterprise 14:01
The speaker says it is easier to start with a strong product and then make it work for enterprise needs than to begin with compliance and admin rules and try to build a great experience later. He says the team focused on product quality first, while keeping data security and privacy solid. That gave them a base they could grow from.
Agents and cheap tokens 17:30
He says the market will grow as people see consumer-grade AI at work and expect the same at the office. The near-term path is to win through email, calendar, and meeting help, then earn the right to connect CRM and other systems. He also says cheaper tokens will make more small tasks worth automating, from prep notes to checking restaurant wholesale prices.
Product choice and market learning 28:30
Town avoids being boxed into one narrow use case. The idea is to learn from thousands of businesses, see what users actually do, and let the market define which verticals matter. For now, the product aims at broad work tasks, then narrows into specific routines for groups like restaurant owners, VCs, or electricians. The model is to start with a few good automations, then keep expanding them as AI and user interviews improve the library.
Why neutrality matters 34:01
Town says the real edge is not model quality alone. It is neutrality, data ownership, and less dependence on any one lab. Customers should be able to swap model providers, keep their context and memories, and avoid being locked in by custom FDE work. Town also trusts AI to route requests inside a company, so people do not have to wait on Slack or Teams. That is meant to make companies move at AI speed.
Business AI reaches outward 43:00
The speaker says AI for businesses is bigger than coding, but the shift will not hit every company at once. In the near term, coding still feels urgent. For broader business use, the change looks more like the iPhone taking over from BlackBerry. The market is already there, but it will move faster over the next 12 months.
Tinkerers make it work 45:00
Use spreads in odd corners of the economy when someone inside the business is willing to tinker. A surfboard rental shop used the product to compare pictures of boards and beach buggies, then decide whether to return a deposit. The real blocker is not model quality. It is that most people are not tinkerers, so the product has to show value and do the work for them.
Agents and ad pressure 57:00
Amazon may need a separate, ad-filled experience instead of blocking agents, because a headless Amazon could still be huge. But AI is moving so fast that no one can predict which platforms win, and Amazon could lose part of its ad business trying to fight that shift.
AI-generated summary. It can be wrong or incomplete - check anything that matters against the original.
