@Vtrivedy10 agree that observability is a key mechanism for continual learning, but it's limited to hillclimbing for model / agent builders. the average agent user lacks such signals and needs agents that learn from their own experience
the core of this โ knowledge that compounds across conversations, health checks, markdown โwikiโ with backlinks โ is how Letta agents have been managing their memory for ages ๐
the difference is your agent builds and maintains it all. you just talk to it.
> an agent whose identity lives in its memory, not its model weights
> differentiator for [enterprise] agents will increasingly be what memory they have accumulated rather than which model they call
As AI reasoning gets good enough, we think memory will be the next bottleneck for agents. Can your agent improve with more experience?
We call this Memory Scaling, and it's related but different from continual learning. A few examples and challenges:
https://t.co/raIa0U7MPs
The new Anthropic managed agents API is basically the Letta API that we've had since a year ago, but closed source and with provider lock-in. They even have read-only memory blocks and memory block sharing -- something which was unique to the Letta agents for a long time.
Funny enough, we actually don't think this is the direction agents are going to go. Having API interfaces for memory blocks and tools is certainly convenient - you can spin up stateful agents as API services with just a few lines of code. But its also limiting: LLMs today are extremely adept at computer-use, and representing their memories in this way limits the action space of agents and their ability to learn.
It's important to remember that just because something comes out of a frontier lab, doesn't mean its the "right" answer long-term. The Letta API ~1 year ago was somewhat of an antipattern in a sea of agent framework libraries offered by every lab. But now, stateful agent APIs are becoming the new norm - especially as providers try to lock in memory/state into their platforms to increase switching costs (which is exactly why we believe memory should live outside of model providers)
If you want to see what the future is going to look like, follow @Letta_AI