After years of pouring billions into anything labeled AI, investors have started drawing a hard line. TechCrunch asked VCs at 645 Ventures, F Prime, AltaIR Capital and Emergence Capital what they are no longer looking for in AI SaaS startups. The answers converge on a single idea: if an AI agent can already do it, or if it has no proprietary data underneath, it is no longer investable. That verdict is a useful lens for evaluating the GEO and AI-visibility tools now flooding the market.
Key takeaways
- 645 Ventures managing partner Aaron Holiday now calls a whole set of categories "quite boring": thin workflow layers, generic horizontal tools, light product management and surface-level analytics. The common thread is "anything an AI agent can now do." - F Prime's Abdul Abdirahman adds that generic vertical software "without proprietary data moats" has fallen out of favor. - AltaIR Capital's Igor Ryabenky goes further: if your differentiation rests mainly on UI and automation, that is no longer enough, because the barrier to entry has dropped and real moats are harder to build. - The GEO implication: a tool that only sends prompts to a model and charts the output is exactly the "thin workflow layer" investors are walking away from. Proprietary, category-level data is the moat that survives. - Two tests now decide durability: does it own data an agent cannot regenerate, and does it live where agents work rather than only behind a web dashboard.
What investors are walking away from
Holiday's list reads like an obituary for the easy AI startup. Thin workflow layers, generic horizontal tools, light product management, surface-level analytics, all boring now, because an agent can do the task on demand. When the marginal cost of a capable agent action approaches zero, wrapping that action in a subscription is not a business.
Abdirahman sharpens it around data. Generic vertical software without a proprietary data moat is out. The software can look polished, but if the underlying data is public or trivially reproducible, there is nothing to defend. Ryabenky closes the loop: UI and automation as your core differentiation no longer clears the bar, because the entry barrier collapsed and true moats got harder to build. Polish is table stakes, not a moat.
The pattern underneath
Strip these views down and one requirement remains: own something an agent cannot recreate on demand. That is almost always proprietary data, a dataset gathered, structured and maintained in a way a competitor, or a coding agent spun up in a weekend, cannot simply reproduce. Everything else, the interface, the automations, the workflow glue, is now cheap to build and therefore cheap to copy.
What this means for GEO
Apply the same test to the GEO tool market and it sorts quickly. A large share of AI-visibility products are, in this framing, thin workflow layers: fire a set of prompts at ChatGPT, count brand mentions, render a dashboard. That is precisely the category investors just labeled boring, because an agent can do it and there is no proprietary data underneath. If a competitor, or a user with a coding agent, can rebuild your tool in a weekend, you are selling automation, not a moat.
