Embedded AI - Intelligence at the Deep Edge

Send us Fan Mail

In this episode, we explore why today’s AI agents, despite huge advances, still fall short of true, unsupervised autonomy. We break down the findings from two key studies that map out both the architectural and operational barriers to fully independent systems.

First, we describe the four levels of agent autonomy, showing why most current systems are stuck at Level 2 or 3. We discuss deep technical constraints — short memory windows, unreliable vector databases, and reasoning gaps that cripple long-term planning and common sense.

Then, we turn to the real-world production limits: missing authentication flows for headless agents, lack of durable schedulers for recurring work, and security threats like prompt injection and non-deterministic outputs.

The takeaway? True “Level 4 autonomy” remains out of reach. The near future of reliable AI will depend on human-in-the-loop systems — smarter collaboration between people and agents, not full independence.

Support the show

If you are interested in learning more then please subscribe to the podcast or head over to https://medium.com/@reefwing, where there is lots more content on AI, IoT, robotics, drones, and development. To support us in bringing you this material, you can buy me a coffee or just provide feedback. We love feedback!

Send us Fan Mail

In a move that's sending shockwaves through the maker community, Qualcomm has officially acquired Arduino. In this episode, we look at one of the biggest tech stories of October 2025.

First, we explore the "why" behind this deal. While Arduino democratized electronics, its core technology was falling behind modern 32-bit competitors. We discuss how this acquisition is Qualcomm's strategic play to dominate the future of IoT and the intelligent edge.

Then, we get our hands on the main event: the brand-new Arduino UNO Q. This isn't your old hobby board. We'll detail its "dual-brain" architecture, which pairs a powerful, Linux-capable Qualcomm quad-core processor with a real-time STMicroelectronics chip, effectively transforming the UNO from a simple microcontroller into a complex Single-Board Computer.

Finally, we tackle the elephant in the room: the future of open source. Qualcomm promises independence, but history is littered with cautionary tales like MakerBot and Nest. Join us as we analyze the potential risks and rewards of this industry-shaking merger.

Support the show

If you are interested in learning more then please subscribe to the podcast or head over to https://medium.com/@reefwing, where there is lots more content on AI, IoT, robotics, drones, and development. To support us in bringing you this material, you can buy me a coffee or just provide feedback. We love feedback!

Send us Fan Mail

Is your business prepared for Australia's shift from voluntary AI ethics to mandatory regulation? In this episode, we unpack a comprehensive guide for establishing a robust AI governance framework within large Australian enterprises. Learn about the multi-layered governance structure, the five core policies you need to implement, and how to create a tiered training program that gets your entire organisation on board. Tune in to proactively manage legal, ethical, and reputational risks and align your AI strategy with national and global standards.

Support the show

If you are interested in learning more then please subscribe to the podcast or head over to https://medium.com/@reefwing, where there is lots more content on AI, IoT, robotics, drones, and development. To support us in bringing you this material, you can buy me a coffee or just provide feedback. We love feedback!

Send us Fan Mail

In this episode, we ask: Is the rise of Artificial Intelligence as pivotal as the invention of electricity or the internet? We look at this history-shifting moment by tracing AI's evolution from the first "perceptron" in the 1950s, through the infamous "AI Winter," to today's deep learning revolution.

We then peek under the hood of the Large Language Models (LLMs) powering our world, breaking down core concepts like the attention mechanism and temperature. We also confront their biggest weaknesses, including hallucinations and limited memory.

Finally, we get practical, exploring how corporations are using AI for everything from contract review to advanced knowledge management and automation. But it's not all about opportunity—we also tackle the critical risks, ethical concerns, and governance challenges, from cultural bias to data security, that we must navigate.

Support the show

If you are interested in learning more then please subscribe to the podcast or head over to https://medium.com/@reefwing, where there is lots more content on AI, IoT, robotics, drones, and development. To support us in bringing you this material, you can buy me a coffee or just provide feedback. We love feedback!

Send us Fan Mail

In this episode, we explore why autonomous AI agents are being called the next frontier for business. 

We're moving beyond isolated AI projects and into the era of the "agentic enterprise," where intelligent, autonomous agents are woven directly into the fabric of daily operations. But what exactly is an AI agent? We break down their key traits—autonomy, adaptability, and goal-orientation—and explain how they differ from older automation tools like RPA by tackling complex thinking, not just repetitive tasks.

Join us as we discuss the massive potential for agents to revolutionize everything from customer service and finance to IT, driving faster resolutions and higher conversion rates. We also confront the critical hurdles to adoption, including technical reliability, systems integration, and new governance risks. Finally, we'll explain why winning with AI agents isn't just about deploying new tech; it's about fundamentally reimagining workflows and building a solid foundation of data and governance.

Support the show

If you are interested in learning more then please subscribe to the podcast or head over to https://medium.com/@reefwing, where there is lots more content on AI, IoT, robotics, drones, and development. To support us in bringing you this material, you can buy me a coffee or just provide feedback. We love feedback!

Send us Fan Mail

In this episode, we dive into one of the biggest challenges in AI today, bias in large language models. The research shows that these models don’t represent all of humanity; instead, they overwhelmingly reflect the values and psychology of Western, Educated, Industrialized, Rich, and Democratic, or “WEIRD”, societies. 

One source frames this as more than just an ethical problem. It’s a real business risk, with potential for legal exposure and reputational damage. Worse still, alignment techniques often create what’s been called an “alignment veneer,” making systems look safe while masking deeper, systemic biases. The other source puts numbers behind this concern, showing through cross-cultural data that LLMs mirror WEIRD thinking styles, essentially becoming parrots of a narrow slice of humanity. Together, they make the case that both businesses and researchers need to go beyond surface-level fixes and build governance models that tackle the cultural imbalance at the core of these systems.

Support the show

If you are interested in learning more then please subscribe to the podcast or head over to https://medium.com/@reefwing, where there is lots more content on AI, IoT, robotics, drones, and development. To support us in bringing you this material, you can buy me a coffee or just provide feedback. We love feedback!

Send us Fan Mail

In this episode, we investigate the growing conversation around deterministic large language models (LLMs), models designed to always return the same output for the same input. We contrast this with the more common stochastic LLMs that rely on random sampling and parallel computation, making their outputs variable even with identical prompts.

We explore the engineering efforts aimed at reducing this variability, including recent claims by Thinking Machines about “batch invariance.” While it’s a step forward, some analysts argue it’s overstated as a singular breakthrough. The episode dives into the multiple causes of non-determinism, from floating-point arithmetic and system-level batching to architectural features like Mixture-of-Experts.

We also weigh the pros and cons of determinism. On the plus side: improved debugging, reproducible benchmarks, and greater trust in high-stakes applications like finance or medicine. On the downside: reduced creative output, increased computational overhead, and significant engineering complexity.

Ultimately, we ask: Is true end-to-end determinism a worthwhile goal—or just an ideal that forces too many trade-offs?

Support the show

If you are interested in learning more then please subscribe to the podcast or head over to https://medium.com/@reefwing, where there is lots more content on AI, IoT, robotics, drones, and development. To support us in bringing you this material, you can buy me a coffee or just provide feedback. We love feedback!

Send us Fan Mail

In this episode, we explore the emerging reality of a "generative AI plateau." For years, the path to better AI has been a simple one: bigger models, more data, and more compute. But now, that brute-force approach is showing diminishing returns. We'll discuss why the industry is hitting this wall, what new strategies are emerging to break through it, and what this all means for the future of AI and the global economy.

We'll break down the core reasons for the scaling slowdown, including the exhaustion of high-quality public training data, the astronomical costs and environmental impact of massive models, and the fundamental architectural limits of the current Transformer paradigm.

We'll debate whether scaling current models can ever lead to Artificial General Intelligence and explore alternative approaches like "test-time knowledge recombination."

Support the show

If you are interested in learning more then please subscribe to the podcast or head over to https://medium.com/@reefwing, where there is lots more content on AI, IoT, robotics, drones, and development. To support us in bringing you this material, you can buy me a coffee or just provide feedback. We love feedback!

Send us Fan Mail

In today's episode, we're tackling one of the biggest questions facing modern businesses: What's the best way to introduce AI tools to your organization? Is it a top-down corporate mandate, or a bottom-up, grassroots movement? We'll explore the pros and cons of both, and reveal what the data shows is the most effective approach.

The future of AI in business isn't a battle between top-down and bottom-up. It's about a symbiotic relationship where leadership provides the vision and employees provide the innovation. It's a journey of democratizing AI, where everyone plays a role in transforming the organization.

Support the show

If you are interested in learning more then please subscribe to the podcast or head over to https://medium.com/@reefwing, where there is lots more content on AI, IoT, robotics, drones, and development. To support us in bringing you this material, you can buy me a coffee or just provide feedback. We love feedback!

Send us Fan Mail

Exploring what it means for a system to converse like a human but forget like a goldfish. Today, we're diving into a topic that's both a technical puzzle and a philosophical mystery: the statelessness of large language models, or LLMs.

Think about the last conversation you had with an AI. It felt real, didn't it? It seemed to understand you, to reason, and to respond. But what if I told you that in the very next moment, it completely forgot everything you said? This is the core paradox we're tackling. These models, which can talk like a human, have no persistent memory. They live in an eternal present, forgetting their past like a goldfish.

The title "Ghost in the Machine" is a nod to a famous philosophical concept, but we're flipping it on its head. The original idea was a critique of the human mind, but in the world of AI, the ghost—the illusion of consciousness—is the conversation itself, while the machine behind it is an empty vessel with no past.

So, how does this work? We'll break down the technical magic, from the limited "context window" that acts as the AI's short-term memory to the sophisticated external systems like Retrieval-Augmented Generation (RAG) and vector databases that developers are using to give these models a kind of artificial, long-term memory.

But this isn't just a technical discussion. The statelessness of LLMs has profound ethical and safety implications. How can we hold a system accountable for its decisions if it can't remember its past actions? And how do we tackle issues like bias when the model is unable to learn from its own mistakes?

Join us as we explore the future of stateful AI agents, the new class of models that can remember and learn. We'll examine the promise of these more capable systems, as well as the new risks they introduce, all while asking the big question: what does it mean to be a partner with a mind that's both brilliant and amnesiac?

Support the show

If you are interested in learning more then please subscribe to the podcast or head over to https://medium.com/@reefwing, where there is lots more content on AI, IoT, robotics, drones, and development. To support us in bringing you this material, you can buy me a coffee or just provide feedback. We love feedback!