Two years ago, every agent framework invented its own vocabulary. Today, the shape is converging. Most serious frameworks now agree on the same primitives: a way to call tools, a loop that decides what to do next, and some form of shared state.
That convergence is good news. It means the hard part is no longer learning a framework. It is designing the tools, the loops, and the state in a way that holds up under real use.
What I expect next
- Shared schemas for tool calls
- Standard ways to inspect agent state
- Portable checkpoints across frameworks
- Better defaults for verification loops
The frameworks will keep moving. But the interesting work is shifting from how to build an agent to how to keep one honest at scale.
