the ultimate moat: relationship
As context lengths expand, LLMs are getting longer memories. Voice APIs are sounding more human. Put these together, and you've got AI that can chat with you like an old friend, not a stateless machine.
Imagine an AI that's been your business coach for years. It knows your whole career story - your wins, your setbacks, that time you almost quit but didn't. When it gives you advice, it's not generic. It's tailored to you, because it knows you.
The tech to make this happen is surprisingly simple. In many circumstances just an append-only text file will do. But that file is gold. It's the record of your relationship with the AI. AI startups aren't going to let users export these files easily. They're too valuable. They're the moat that'll keep users from jumping ship.
Startups will soon only be able to compete at the application layer. The big tech giants will dominate the cutting-edge frontier models by winning at the energy and compute infrastructure level, leaving startups working on foundation models in the dust. So the competition at the application layer will be intense. If the frontier model powering these applications is the same, what do these startups have to retain users? The relationship.
Once you've spent months or years talking to an AI coach or therapist, switching to a new one will feel like starting over. The switching cost, perhaps the only power that these AI coaching startups will have, is huge.
We're going to see this play out soon in areas where you'd normally build a relationship with a human expert. AI therapists might be ethically tricky, but AI business coaches? Relationship advisors? They're coming.
The companies that nail this will have users for life. Because who wants to switch coaches halfway through a promotion cycle or halfway through a messy moment in a relationship?