In this Christmas episode of the AI-Shed podcast, Alessandro and Giovanni step away from predictions and buzzwords to do what engineers actually do at the end of the year: a systems review.

They look back at what really worked in AI over the past year, where expectations clearly outran reality, and why many so-called “agentic” systems failed in very predictable ways. From overconfident automation disguised as autonomy to agents that optimized themselves straight into chaos, the episode unpacks why context, governance, and clear objectives mattered more than raw intelligence.

Rather than focusing on demos or theoretical promises, the discussion highlights a deeper shift that happened this year: a change in mindset. Engineers and decision-makers are no longer just asking whether something can be automated, but whether it should be allowed to decide at all.

The episode closes with practical, low-pressure recommendations for the holidays: how to learn about agentic AI without hype, why slowing down and revisiting systems thinking matters, and how small, safe experiments can teach more than ambitious prototypes.

A calm, slightly irreverent end-of-year debrief—designed to help listeners start the next year with fewer illusions and better architectures.

In this episode of the AI-Shed Agentic AI Podcast, Alessandro and Giovanni explore why agentic AI cannot be understood as a single intelligent component.

They explain how the moment an AI system begins to act with agency, it becomes part of a wider System of Systems—interacting with humans, data sources, tools, rules, and feedback loops. Through everyday examples and practical analogies, they show that behavior emerges from interactions, not from intelligence alone.

The conversation highlights why many AI failures are not model failures but system and interface failures, and why introducing an agent without redesigning responsibility, authority, and context leads to slow but serious organizational risks.

This episode offers a systems-engineering perspective on agentic AI, helping listeners move beyond hype and toward a more realistic, resilient way of designing and deploying intelligent systems.

In this episode, Alessandro and Giovanni explore why explainability is no longer enough for modern AI and agentic systems—and why observability must take center stage. They explain that explainability is retrospective (“why the system thinks it acted”), whereas observability provides real-time insight into what the system is doing, whether it remains within its safety envelope, and how far it is from violating constraints.

Drawing parallels with fly-by-wire and safety-critical software, they show that autonomy increases—not reduces—the need for instrumentation, logging, monitoring, and traceable reasoning. The conversation emphasizes that trustworthy agentic AI requires continuous telemetry, drift detection, guardrail activation logs, and visibility into planning and sub-goal generation.

To illustrate the risks of missing explainability, they recount a striking real-world example: a Boeing 737 automated braking system that behaved “perfectly” according to its internal logic but offered no cues to pilots. This opacity led to confusing and unsafe events until engineers added simple cockpit messages. The system didn’t need new logic; it needed to communicate.

The core message of the episode is clear:
 Autonomous systems cannot be trusted unless their behaviour is continuously observable. Explainability is helpful, but observability is essential for safety, certification, and human-machine collaboration.

In this episode of Agentic AI, Alessandro and Giovanni explore why robust systems engineering remains essential in the age of machine learning and agentic AI. Through a technical yet conversational exchange, they revisit the Therac-25 accident as a cautionary tale demonstrating how inadequate hazard analysis, missing interlocks, and reused software can lead to catastrophic failures. The discussion highlights that similar risks now arise with modern autonomous and learning-based systems.

They explain that the biggest change introduced by machine learning is not the algorithms themselves, but the entire data management process, dataset governance, and the impact of stochastic behaviour on certification. Current aerospace and safety-critical standards cannot yet certify learning systems because they adapt without explicit programming, making traceability and deterministic verification difficult.

The episode connects this to broader system integrity concerns: command-and-control links must be highly reliable, secure, and resistant to jamming or spoofing. Drawing parallels with fly-by-wire systems, Giovanni stresses that communication integrity, safety-risk mitigation, and a solid architectural “required base” are mandatory for high-complexity system development.

Alessandro and Giovanni conclude that agentic AI—systems capable of autonomous reasoning and decision-making—must be engineered within strict constraints: robust data governance, validation loops, safety monitors, and lifecycle discipline. Autonomy does not replace systems engineering; it makes it indispensable. The key message: without structure, autonomy becomes instability, and past failures like Therac-25 must guide the design of tomorrow’s intelligent systems.

Episode 1 — “What Is Agentic AI, and Who Are We?”

Episode 1 introduces the world of AI-Shed and the core idea behind the series: agentic AI—systems that don’t just react, but decide, adapt, and collaborate with humans.

Alessandro opens by explaining AI-Shed as a creative hangar: a meeting point for engineers, designers, philosophers, and curious minds. The goal is to explore how intelligent systems are reshaping the way society builds, decides, and imagines the future. The origin of the name AI-Shed is presented with humor: if Apple had a garage, AI-Shed has a shed.

He then welcomes Giovanni, long-time collaborator and systems engineer. Together, they define agentic AI as the shift from automation to autonomy: systems with goals, environmental awareness, and initiative. Examples range from fictional companions like JARVIS to real-world technologies such as predictive logistics systems and adaptive digital twins. They link this type of intelligence to Bayesian reasoning, uncertainty management, and systems engineering principles.

Throughout the discussion, they highlight how science fiction has shaped engineers’ imagination—from Her and Ex Machina to WALL-E—not because fiction predicts reality, but because it reveals what is at stake when machines interact with humans.

The episode also stresses the importance of responsible design, embedding ethics and context into systems from the start rather than as late add-ons.

Before closing, Alessandro and Giovanni show how agentic AI already exists in everyday tools: email assistants that learn patterns, note-taking AI that builds personal knowledge graphs, adaptive workflows that plan tasks autonomously. They encourage listeners to try a small experiment: choose a repetitive task and delegate it to an adaptive AI system rather than a fixed automation.

The episode concludes by linking agentic AI to systems engineering: understanding how autonomous components fit into larger structures. This sets the stage for Episode 2, which will dive into the cautionary tale of Therac-25 and the importance of system-level thinking in AI.

What happens when artificial intelligence stops being a tool and starts acting like a partner? In this first episode of AI with Agency, aerospace engineer Alessandro Migliaccio explores how machines are evolving from passive programs into systems that perceive, decide, and adapt. Drawing parallels from pop culture — from Her and Iron Man to Black Mirror and Blade Runner — the episode examines the thin line between automation and autonomy. Through real-world examples in aerospace, finance, and healthcare, we unpack what “agency” really means in AI — and how it’s quietly redefining the future of work, ethics, and human creativity.