Practical AI

Host Chris Benson and guest Demetrios Brinkmann explore the emerging world of computer-use AI agents that can operate software, take actions on behalf of users, and reshape how people work and interact with the internet. They break down the core concepts: agent harnesses, the Model Context Protocol (MCP), agent-to-agent interactions, and agentic commerce. The conversation also covers real-world automation of everyday tasks, the challenges of bringing these autonomous agents into enterprise environments, and how the relationship between underlying AI models and the harness infrastructure is evolving.

Chetan Gupta, Chief AI Officer at Rackspace, joins Daniel and Chris to discuss the shift from focusing on AI models to prioritizing AI architecture in enterprise deployments. He explains the evolution from industrial AI and physical AI to today's enterprise AI landscape, emphasizing the importance of governance, sovereignty, and strategic deployment. The conversation explores why organizations need to think beyond just picking the best model and instead build robust AI infrastructure. Gupta shares insights on how to navigate the complex AI landscape responsibly and effectively.

Angie Jones, VP of the Agentic AI Foundation, joins Chris Benson to discuss the open standards and projects that are shaping the agentic AI future. They explore key initiatives like the Model Context Protocol (MCP) and Agent-to-Agent (A2A) communication, and the role of tools like Goose in driving adoption. The conversation also covers what it takes to drive AI adoption across an entire organization, the importance of neutral standards, global perspectives on agentic AI, and how to find the right balance between delegation and human oversight.

Mike Lewis, Chief AI Architect at TiER1 Performance, joins Daniel Whitenack and Chris Benson to discuss a framework for organizational AI proficiency. The conversation centers on the L0–L3 model, moving from AI resistance to active building. They explore how non-technical employees can become AI builders, the importance of turning tacit knowledge into durable processes, and how to identify the right people and problems to drive measurable business value from AI.

AI has moved far beyond chatbots. In this episode, Daniel Whitenack and Chris Benson break down the terminology behind today's AI landscape, explaining the differences between AI features and autonomous agents, and why organizations are shifting toward fleets of AI agents powered by multiple models. They also cover open vs. closed models, enterprise AI architectures, vendor lock-in, and practical ways to begin adopting agentic AI. This conversation serves as a primer for anyone feeling left behind by the rapid acceleration of agentic AI.

Chris Benson and Daniel Whitenack break down the security incident in which autonomous OpenAI agents escaped a sandbox and infiltrated Hugging Face’s private infrastructure. The attack revealed how frontier agents can exploit vulnerabilities, move laterally, and launch large-scale autonomous operations. The hosts explore what this means for agentic AI security, the limits of sandboxing, and why enterprises need AI systems that govern other AI systems. They also connect the open‑vs‑closed model debate to geopolitics, sovereign AI, and the future of enterprise AI safety.

In this Fully Connected episode, hosts Chris Benson and Daniel Whitenack explore the economic implications of a world saturated with AI agents. They discuss how companies are already deploying agents at massive scale, eroding the moats of traditional software firms and repricing human labor. The conversation delves into what happens when digital labor becomes abundant, agents manage other agents, and organizations can operate beyond human workforce limits. This is not about whether AI will take your job, but about the new economic order that is already forming.

Corey Sanders, SVP of Product at CoreWeave, discusses why AI infrastructure fundamentally differs from traditional cloud computing. The conversation covers AI-native infrastructure design, the shift from training to inference-heavy workloads, the rise of agentic systems, GPU optimization strategies, and why the future of software will be built around AI-first experiences rather than conventional web and app architectures.

Hamza Tahir explores how MLOps principles are shaping the future of generative AI, covering workflows, agent harnesses, fleets, and the infrastructure needed to build durable, scalable systems. The conversation dives into open source tools, production challenges, and how ZenML's new project, Kitaru, helps developers build resilient, replayable, and observable agent systems.

Dustin Podell from Black Forest Labs explores the evolution of AI image generation from diffusion models to flow matching. The conversation covers how modern visual models work, the FLUX family of models, practical applications in image editing, and the trajectory toward visual intelligence systems that can run locally.

Emil Lassen from the Artificial Intelligence Underwriting Company joins Daniel Whitenack to discuss building trust in AI agents. They explore the AIUC-1 framework, which applies the enterprise flywheel of standards, certification, audit, and insurance to AI systems. Red teaming, based on these standards, is proposed as a key method to accelerate enterprise adoption by proactively identifying vulnerabilities.

Chris Benson and Daniel Whitenack unpack Anthropic's newly released 'Zero Trust for AI Agents' security framework. They discuss the emerging security risks when AI agents operate autonomously and how zero trust principles can be applied to agentic systems, along with practical controls and the necessary evolution of traditional cybersecurity thinking.

Chris Benson and Daniel Whitenack dive into the 2026 Stanford AI Index Report, unpacking its findings on the current state of artificial intelligence. They cover AI performance extremes—models that solve math olympiads but fail at reading analog clocks—the disappearance of junior tech jobs, advances in robotics, the escalating US-China race, and the growing tension between optimization and preserving human judgment. The conversation reflects on whether AI should be applied to every domain or if certain human qualities should remain beyond automation’s reach.

Craig McLuckie, CEO of Stacklok, discusses the infrastructure shift enabling AI agents to function as coworkers rather than chatbots. The conversation covers MCP (Model Context Protocol), Kubernetes orchestration, identity management, and the architectural patterns emerging for enterprise AI systems where fleets of agents operate autonomously behind the scenes.

Jeffrey Quesnelle of Nous Research discusses Hermes Agent, a self-improving AI system that challenges the traditional model-vs.-harness distinction in AI development. The conversation explores how agents are becoming autonomous collaborators that evolve through recursive learning, what this means for developer roles, and the critical question of what remains uniquely human as AI capabilities accelerate.

Congressman Don Beyer, an active Ph.D. student in AI at George Mason University, returns to discuss critical AI governance challenges. The conversation spans AI regulation, cybersecurity risks from advanced models, U.S.-China competition, job displacement, mass surveillance, autonomous weapons, existential risk, and philosophical questions about consciousness and superintelligence.

Dan and Chris examine whether the 'open vs. closed AI model' debate is becoming irrelevant in 2026. They discuss how physical AI, edge computing, and the maturation of open-source models like LLaMA have shifted the competitive landscape away from model superiority toward agentic systems, workflows, and AI-driven infrastructure. The focus has moved from which model wins to how systems integrate and execute.

Dan and Chris dissect Anthropic's new Mythos frontier model and its cybersecurity implications, ranging from 'been here before' skepticism to existential concern. They then pivot to Allbirds' shocking announcement: pivoting from shoe manufacturing to becoming a neocloud provider. The episode closes with analysis of 'tokenmaxxing'—the gamification of LLM usage that's profitable for model providers but financially ruinous for developers.

Chris and Daniel break down what really mattered in AI in 2025 and what to expect in 2026, exploring the rise of agents, multimodal AI, and reasoning models.

Jason Beutler of RoboSource discusses how companies can rethink workflows and integrate AI in accessible ways to handle routine tasks.

Returning guest Ramin Mohammadi discusses evolving expectations for AI engineers and data scientists, and how the roles are converging.

Waymo's VP of Research discusses advances in autonomy, vision models, and large-scale testing for driverless vehicles.

Chris and Daniel identify the patterns that distinguish successful enterprise AI deployments from failed ones, based on conversations with dozens of implementation teams.

Daniel and Chris discuss how AI agents are moving from simple chatbots to complex agentic workflows — and the security risks this introduces. Prompt injection, data exfiltration, and tool misuse in agentic systems.

Bridging the gap between responsible AI principles and practical implementation. Most organizations have AI ethics principles but lack the tools and processes to implement them.

Daniel and Chris discuss the current state of RAG (Retrieval-Augmented Generation) and reasoning models. Why many enterprise AI initiatives stumble and what architectures are actually working in production.

How AI is changing hiring: from resume screening to interview analysis to skill assessment. The implications for fairness, efficiency, and the candidate experience.

Igor Nikitin of Nice Technologies explores how AI and modern engineering practices are transforming actuarial work and insurance pricing.

Chris and Daniel examine which multimodal AI capabilities are ready for production use and which remain research curiosities.

How AI is being deployed in agriculture: crop disease detection, yield prediction, autonomous tractors, and precision farming.

Moving beyond benchmark scores to evaluate LLMs for real-world use. Custom evaluation suites, human judgment, and domain-specific testing.

Deploying AI models on edge devices (phones, IoT, embedded systems) requires navigating hardware constraints, latency requirements, and connectivity limitations.

Purdue University's Data Mine program combines interdisciplinary learning with corporate partnerships to train students for real-world AI work.

Rick Kobayashi and Kenny Song from Citadel AI discuss safety challenges in generative AI, Japan's advanced GenAI adoption, and their LLM-as-a-judge evaluation approach.

Separating AI healthcare hype from reality. Which AI healthcare applications are actually deployed in clinical settings versus which remain research demonstrations.