Introduction
AI startups are at an inflection point: adoption is still thin, yet the upside is massive. Here’s a concise summary of the key questions that will shape products, teams, costs, moats and capital in the coming years.
Context
The piece opens with a “Home Screen Test”: how many truly AI-native apps sit on your phone’s home screen? Today, few beyond well-known LLMs. This signals untapped space: we’ll change how we work and build, not just how we prompt. The author also notes initiatives (e.g., a16z speedrun 6) as signs of the wave ahead.
Key questions for AI startups
The open issues below will guide ecosystem strategy and investment.
Teams: fewer people or more roles?
AI can multiply leverage—one supervisor with 1000 software agents? Maybe. Yet if human “taste” and design remain critical, headcount may still grow to scale well.
Defensibility: what moat in a fast-copy world?
If features commoditize, moats may shift to user growth/network effects or relentless iteration and shipping. Alternatively, pursue multi-year, CapEx-heavy bets (hardware, space) that are hard to clone.
Cost: building is cheap, distribution isn’t
Foundation models are CapEx-heavy, while many apps are easy to build. Real costs cluster in user acquisition: crowded markets and scarce attention.
Organization: traditional functions or convergence?
With multimodal tools turning PRDs/wireframes into software, engineering, product, and design may converge. Work structures evolve from cottages to factories to agents.
Geography: does SF remain the hub?
The Bay Area still has advantages, but decentralization rises as building gets easier and knowledge spreads. Founders and capital may distribute globally.
Venture and stages
If profitable products emerge from tiny teams anywhere, VC may become more distributed and growth-like. Pre-seed/seed/Series A boundaries could blur.
Historical precedents
From workshops to corporations: technology reshapes business structures. With AI, agents, and compute, today’s templates may be insufficient to organize production.
Optimistic view and risks
Best case: few people build more, moats come from tech and features, the ecosystem thrives. But scale may centralize with incumbents holding data centers, data, and compute; they could integrate AI before startups win distribution.
Conclusion
The next few years favor “business logic” atop models: products, UI, and rapid iteration over frontier research. Source: "AI will change how we build startups — but how? We still don't know a lot. A list of the questions."
FAQ
- Why does the Home Screen Test matter in AI research? It exposes the gap between hype and real AI-native adoption, pointing to practical innovation space.
- Will AI startups have smaller teams? It depends: AI boosts output, but human-critical work (e.g., design) can remain bottlenecks.
- What is the strongest moat for AI startups? Either user networks and execution speed, or long-horizon, CapEx-heavy bets that resist cloning.
- Will AI startups be cheaper to build? Building yes, but marketing and distribution stay costly and highly competitive.
- Will the Bay Area lose centrality for AI search and AI models? It may soften, yet SF’s network and capital advantages still matter.
- How will venture change with LLM-driven traffic? More distributed and growth-oriented, with fuzzier boundaries between funding stages.