← Back
Theo July 10, 2026 34s

The Eras of AI Agents

Read full transcript 1 segments
  1. I think of these almost as eras now, I think of these almost as eras now, where Sonnet 3.5 is the tool call era. where Sonnet 3.5 is the tool call era. where Sonnet 3.5 is the tool call era. Not that it was the first model that Not that it was the first model that Not that it was the first model that could do tool calls, rather it was the could do tool calls, rather it was the could do tool calls, rather it was the first one that did them consistently and first one that did them consistently and first one that did them consistently and reliably enough where you could use this reliably enough where you could use this reliably enough where you could use this for day-to-day coding work. Then we got for day-to-day coding work. Then we got for day-to-day coding work. Then we got Opus 4.5, which was able to do much Opus 4.5, which was able to do much Opus 4.5, which was able to do much longer running tasks without losing longer running tasks without losing longer running tasks without losing track of what it's working on. Mythos is track of what it's working on. Mythos is track of what it's working on. Mythos is another jump to orchestration. It feels another jump to orchestration. It feels another jump to orchestration. It feels to me like it's the first model that to me like it's the first model that to me like it's the first model that doesn't just understand your code base, doesn't just understand your code base, doesn't just understand your code base, but it understands itself. And it knows but it understands itself. And it knows but it understands itself. And it knows how to spawn additional models and break how to spawn additional models and break how to spawn additional models and break up work in a way where it can be up work in a way where it can be up work in a way where it can be completed more reliably and then completed more reliably and then completed more reliably and then verified afterwards. And if you tell the verified afterwards. And if you tell the verified afterwards. And if you tell the model to do that, it will just do it.

Summary

The discussion highlights distinct eras in AI model development, focusing on tool call consistency with Sonnet 3.5, long-running task capability with Opus 4.5, and the emergent concept of self-understanding and orchestration with Mythos. The practical takeaway is the evolution of AI from task execution to intelligent self-management and complex workflow handling.

View original episode ↗