Interesting times.
Overarching theme of startups right now is this: software is becoming disposable, and the agent economy (where agents are economic actors) is coming together. Vertical Saas becomes nondefensible since the barrier to coding your own is now just so low. Moats can now be one of three things:
1. Data moat
2. Hardware/infrastructure moat
3. Relationship/infrastructure moat
Basically, just things that agents use but you can't just claude code yourself, which can be things that require hardware support (e.g. iMessage or runtime infrastructure) or relationship support (e.g. payment or carrier relationships)
Data from YC reflects this. Over the past three batches, agent infrastructure (memory, identity, sandboxes, comms, payments) went from 17 percent in W26 to 26 percent in P26 and 25 percent in S26 (as of now). This makes sense, since in the agent economy and the world where people build all software themselves, these are the things riding the wave because they are difficult to build and universally needed.
Robotics and defense are also seeing a steady rise from 8 to 7 to 15 percent. This makes sense because hardware is a moat (and eventually, AI will need to interact physically with humans). Nothing much to talk about here.
What is interesting to me though is the falling number of AI-native services startups, or full-stack startups. I would imagine this to be the percent counter to the failing moat of vertical Saas, since an AI-native accounting firm or AI-native law firm is inherently moated. Also, didn't Corgi, the AI-native insurance firm, just become a unicorn? Perhaps there's something interesting to be thought about here.
Dev tools also saw a huge drop from W26. I guess people are starting to realize that Claude Code and VS/Cursor is probably enough for most coding applications (or in the near future). Sales and GTM agent startups are also gone. Again, Claude can just do this for you.
I think what's happening right now is that we're seeing the software infrastructure wave in the AI cycle, similar to that of the Dot Com cycle. First, there were hardware infrastructure startups. Then it moved to software infrastructure, before finally going to the application layer. The big question I'm thinking about right now is whether there will still be unicorns in the application wave, or will it just be independent builders + builders in legacy + relationship-based giants building internal software.
Interesting things to ponder on, for sure.