By end of 2025, it should mature as a library where developers are comfortably using it to build llm applications with confidence.

• The best library to build and auto-optimize agents [including tracing, production-grade, design patterns,

For developers

The current documentation can be intimidating for normal users. And the general feedback is our learning curve is higher than Dspy, here are things we can do:

1. Docs

1. New landing page + onboarding video ‣
2. change the tutorials → developer notes (mainly for contributors or deep dive)
3. start new tutorials with simple code demonstrations and colabs.
4. improve the demonstrations on integrations.
5. create tutorials on using optimizers.
6. blog on creating datasets for RAG.

2. Engineering

1. support pydantic as data modeling too, besides of with our customized DataClass which was for more simpler and more controlled use cases.
   1. Whatever functionality with DataClass we will support with BaseModel.
   2. PR here: https://github.com/SylphAI-Inc/AdalFlow/pull/383
2. improve the tracing+ diagnose to help users debug.
   1. should we build ourselves or integrate agent observability library like agentops
3. Add support for tools (tools center) , such as
   1. web search [already has first version]
   2. code-gen + execution
      1. support front end code/python/java code and be able to execute the code to plot images or generate various response
      2. might need to consider sandbox
      3. but more research should be done and comparing other libraries and learn the best
   3. ensure the community can integrate them
   4. compute-use [beyond api]
   5. MCP integration [WIP June 2, 2025]
   6. tool management
4. Better design of the apis (such as call, forward, bicall) based on feedback
5. Build a runner and improve the agent system
6. Support stream_async, async call better.