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 [evaluated on 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. start new tutorials with simple code demonstrations and colabs. The older tutorials, renamed as developer notes, are mainly for contributors or deep dive
3. improve the demonstrations on integrations.
4. create tutorials on using optimizers.
5. blog on creating datasets for RAG.
6. blog on creating datasets for agents

• [ ] New Tool/MCP tutorial June 22, 2025 [Li and Junyuan]
• [ ] New Agent tutorial [Jinha and Li]

2. Engineering

1. user reported issues [always on]
2. model-client
   1. support reasoning models: https://platform.openai.com/docs/guides/reasoning?api-mode=responses
3. 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
4. improve the tracing+ diagnose to help users debug.
   1. should we build ourselves or integrate agent observability library like agentops