Last Friday (14th), I attended an event in San Francisco organized by Deeplearning.ai — the well-known learning platform led by Andrew Ng — focused on AI, Agents, and the broader ecosystem: AI Dev 25.

The goal of the event was to create a more vendor-neutral space for AI developers to connect — and it worked!

As expected, the event was highly technical, but with a strong emphasis on real-world use cases.

The main discussions revolved around AI Agents and their applications.

Talks from CrewAI and LangChain stole the spotlight, bringing a lot of clarity to attendees. The format was heavily focused on live coding and demos, with very few PPTs.

Andrew Ng was very vocal about pushing back against the hype that “AI will write all code by the end of the year.” Instead, he reinforced the classic building blocks approach — AI Agents, LLMs, workflows, etc., are just more tools to help solve problems.

My key takeaways:

1. CrewAI: If you’re thinking about building AI Agents, CrewAI is incredibly intuitive. If you’ve used Django before, the development flow and project organization feel very similar.

2. Groq: Incredible performance! I saw a conversational AI Agent running inference with gpt-4o-mini from OpenAI, using Groq Cloud, and the end-user experience was real-time.

3. MCP: It’s becoming the standard for connecting AI Agents -> Tools. However, Agent -> Agent communication is still an open question — Nvidia is trying to fill this gap with AgentIQ.

4. Long-term Memory: Still in its early stages for practical use cases. This aligns with my previous perception that few real-world scenarios truly require it yet. This was reinforced in a talk by Harrison Chase, CEO & Co-founder of LangChain.

5. AI Agents Marketplace: Apify seems to be pivoting into a marketplace for Agents and Tools, even with a fair-share model. You can publish your AI Agent or Tool and get paid for it. It’s already integrated with MCP as well.

Across all the talks I attended, AI infrastructure was highlighted as a critical topic.

Connecting this to Nvidia’s GTC (where I am now), it’s clear that AI infrastructure (especially on-prem) is a major area of focus and investment.

And finally, as I have been saying for a while, AI Agents are just intelligent workflows — something we’ve been doing for decades. The difference? We now have smarter, more versatile building blocks to work with.