Where do AI agent frameworks add value?
Two founders emphasize different parts of the problem: operating the agent stack and improving the context it uses. These positions can coexist.
One view
You need the whole agent stack — it is many components, not one tool.
João Moura, CrewAI EP 29 · The Emerging AI Agent Stack | CrewAI’s João MouraCompanies will need a suite … because of the stack … is many different things … not only the agents … the tool, the authentication, the authorizations, and everything's in between.
The other
As abstractions stabilize, production context still needs work.
Jerry Liu, LlamaIndex EP 61 · The AI Framework Era Is Over: Why Context Is the Moat | Jerry Liu of LlamaIndexthe abstractions were solidifying, but the need for high-quality production context was still evolving.
Beyond the podcast
Anthropic's guidance on building agents found that the most successful implementations it saw used simple, composable patterns rather than complex frameworks. Frameworks simplify low-level work such as calling models, defining tools and chaining calls, but they add layers of abstraction that can hide the underlying prompts and responses and make systems harder to debug; Anthropic suggests starting with model APIs directly, and if you do use a framework, understanding the code underneath (Anthropic, Building effective agents). Anthropic has since noted that much of the tooling it described has changed.
Harrison Chase of LangChain makes the framework case on the same ground Liu does: the hard part of a reliable agent is giving the model the right context at each step. He agrees that agent abstractions can get in the way of that, and argues that a framework earns its place through orchestration and production features such as short- and long-term memory, human-in-the-loop approval and streaming, which a team can also build itself (LangChain, How to think about agent frameworks).
Do you need an agent framework? and AI agent frameworks compared go further, and what is context in AI agents? covers the context problem both sides point to.
More debates
- Will AI replace software engineers? One guest says the agent now writes the code. Another says writing code was never the job.
- Is the agentic AI boom a bubble? Spicy take versus the long view on whether agents are hype or here to stay.
- Do you need to fine-tune a model? Habib says Writer's models spare customers fine-tuning. Srinivasan says most open-model use cases she sees involve fine-tuning.
- Should AI companies build in the open? One guest hopes more companies take an open source route with AI. Another believes in open source but says it is hard to turn into a profitable business.