Demetrios chats with Paul van der Boor and Floris Fok about the real-world challenges of deploying AI agents across @ProsusGroup of companies. They break down the evolution from simple LLMs to fully interactive systems, tackling scale, UX, and the harsh lessons from failed projects. Packed with insights on what works (and what doesn’t), this episode is a must-listen for anyone serious about AI in production.

As Large Language Models (LLMs) evolve, the challenge shifts from raw capability to structuring them into reliable, scalable systems. Many real-world AI products struggle with robustness, complexity management, and evaluation—especially in enterprise contexts. This talk explores how multi-agent systems can help overcome these obstacles by decomposing large monolithic agents into specialized subagents working together in structured architectures. We’ll cover: - Why enterprises struggle to integrate LLM agents effectively. - How multi-agent architectures (Assembly Line, Call Center, and Manager-Worker) improve scalability, modularity, and reliability. - Practical trade-offs and implementation strategies from real-world applications. (planning to adapt my post https://blog.sshh.io/p/building-multi-agent-systems)

Artificial Intelligence is transforming the way we interact with technology, and Agentic AI—systems that exhibit autonomy, adaptability, and decision-making capabilities—is at the forefront of this revolution. But what does this mean for the Arabic language, one of the richest and most complex languages in the world? As we advance AI-driven agents, ensuring they understand, process, and generate Arabic with the same fluency and nuance as English or other dominant languages is not just a technological challenge but a cultural imperative. In this speech, we will explore how Agentic AI can empower Arabic speakers, enhance accessibility, and preserve the linguistic heritage of over 400 million people while driving innovation across industries. The future of AI is agentic. The future of Arabic in AI depends on how we shape it today. Artificial Intelligence is transforming the way we interact with technology, and Agentic AI—systems that exhibit autonomy, adaptability, and decision-making capabilities—is at the forefront of this revolution. But what does this mean for the Arabic language, one of the richest and most complex languages in the world? As we advance AI-driven agents, ensuring they understand, process, and generate Arabic with the same fluency and nuance as English or other dominant languages is not just a technological challenge but a cultural imperative. In this speech, we will explore how Agentic AI can empower Arabic speakers, enhance accessibility, and preserve the linguistic heritage of over 400 million people while driving innovation across industries. The future of AI is agentic. The future of Arabic in AI depends on how we shape it today.

In this episode, we talk with Kenny Daniel, founder of Hyperparam, to explore why actually looking at your data is the most high-leverage move you can make for building state-of-the-art models. It used to be that the first step of data science was to get familiar with your data. However, as modern LLM datasets have gotten larger, dataset exploration tools have not kept up. Kenny makes the case that user interfaces have been under-appreciated in the Python-centric world of AI, and new tools are needed to enable advances in machine learning. Our conversation also dives into new methods of using LLM models themselves to assist data engineers in actually looking at their data.

While online resources make AI knowledge accessible, in-person meetups offer something unique—real conversations, direct access to experts, and unfiltered insights from practitioners. MLOps Days NYC, hosted with JFrog, brings together AI and MLOps professionals for practical talks, networking, and firsthand learning. Join us on March 4th at Rockefeller Plaza for an evening of deep technical discussions and valuable connections.

There seems to be a shift from workflow languages to code - mostly annotation pythons - happening and getting us. It is a symptom of how complex workflow orchestration has gotten. Is it a dominant trend or will we cycle back to “DAG specifications”? At Stitchfix, we had our own DSL that “compiled” into airflow DAGs and at MicroByre, we used a external workflow langauge. Both had a batch task executor on K8s but at MicroByre, we had human and robot in the loop workflows.

This survey examines popular workflow orchestration systems, highlighting their strengths, weaknesses, and key features. It provides insights into how these tools handle automation, scalability, and reliability, helping teams choose the right solution for their needs.

In this MLOps Community Podcast episode, Willem Pienaar, CTO of Cleric, breaks down how they built an autonomous AI SRE that helps engineering teams diagnose production issues. We explore how Cleric builds knowledge graphs for system understanding, and uses existing tools/systems during investigations. We also get into some gnarly challenges around memory, tool integration, and evaluation frameworks, and some lessons learned from deploying to engineering teams.