Building Intelligent AI Agents with Open Source LLMs | smolagents

[@prashlovessamosa]

Man you don't know how good you are please keep teaching.

[@solyarisoftware]

Hi Adam, Thank you for another excellent and comprehensive video; it's greatly appreciated. I'm exploring the development of conversational agentic systems, which I refer to as "Conversation Routines"-a term inspired by OpenAI's Swarm framework. In this approach, business stakeholders (such as domain experts) craft applications using plain natural language instructions. These instructions can invoke functions (utilizing a tool-calling strategy) developed by a separate team of software engineers. I've detailed this concept in my paper, "Conversation Routines: A Prompt Engineering Framework for Task-Oriented Dialog Systems". My critique of smolagents is that the integration of the instruction component (the prompt) with the function logic seems to conflict with the principle of separation of responsibilities between application developers and backend function developers. This blending could lead to challenges in troubleshooting, as the logic is partially embedded within prompt instructions and partially executed at runtime by the LLM in standard programming languages like Python. Nonetheless, I recognize the performance improvements that smolagents offer. What are your thoughts on this? Perhaps it warrants a comparative discussion in a future video :) Thanks again, Giorgio

[@Deeneeshsu]

You know what? Before watching full video my hands automatically goes to like button man :) That much trust I have on your content buddy. Keep it up

[@chronicallychill9979]

I really appreciate your no nonsense delivery and focus on important things for actual practicians. Thanks again for another good one.

[@AdamLucek]

Thanks! For the practitioners is my goal 😃

[@sudhitpanchal4996]

I got this in my google feed, getting better at algo. Will watch after a few hours..

[@Yelonek1986]

Your channel caught my attention because of an Ł in your name. I thought you were Polish, but your accent sounds like you are USA born and raised. I love the content. Very informative and in depth. I can't believe you are not some ML wizz in the Silicon Valley. Keep up the good work!

[@ML-np3jp]

Attractive thumbnail! , 🇮🇳♥️

[@lavamonkeymc]

Amazing video of smolagents! The code agent w function calling is interesting but doesn’t provide full data validation ,almost moving a step backward from thr structured outputs benefits we’ve seen .. For your next video can u look into fine tuning an open source Deepseek with RL GRPO it would be really interesting to see that side of fine tuning since you’ve done a bit before

[@kunalb1995]

Love the thumbnail!

[@AdamLucek]

Thank you sir!

[@attilavass6935]

Definitely going through your code example, great content! Is the code written by Code Agent executed in a sandbox environment like E2B? How about this security issue?

[@AdamLucek]

Yes smolagents has an E2B integration huggingface.co/docs/smolagents/en/tutorials/secure_code_execution It defaults to a local interpreter, which I demonstrated as the concept is the same regardless! There’s more details in the docs link