DiscoverMachine Dreams
Machine Dreams
Claim Ownership

Machine Dreams

Author: Kolawole Samuel Adebayo & Leah Stern

Subscribed: 1Played: 1
Share

Description

Machine Dreams is a weekly dive into the real world of AI. Hosts Kolawole Samuel Adebayo and Leah Stern talk with the founders, researchers, investors, engineers, and stakeholders shaping the future of intelligent machines. These are unfiltered, human conversations about what’s working in AI, what isn’t, and what might reshape the world next. From databases and infrastructure to agents, regulation, the future of work, and more, Machine Dreams brings clarity to arguably the most important tech story of our time.

39 Episodes
Reverse
AI agents are getting easier to build, and businesses are finding more for them to do. But trusting them is a different problem. Some companies are actively blocking agents from their systems, including JP Morgan and Delta, while others build rules so customers can use them safely. This week on Machine Dreams, Sam Adebayo and Leah Stern sit down with Brandon Klein, founder and executive chairman of Collaboration AI (https://collaboration.ai/), a company that helps organizations bring people and ideas together to solve complex problems. Brandon is an engineer who has worked on innovation projects involving NASA and the US government, and he is a strong advocate for what he calls human-first AI.His definition is practical. Human-first means understanding where you want to go with the AI's help and staying at the center of control. His example is his own teenagers using AI that just gives the answer, so they learn nothing. He also explains why two banks can respond to agents in completely opposite ways, one blocking them and one embracing them, and why that difference is culture rather than technology. The centerpiece is a story from his work with the US Air Force. There had never been a proper female flight suit despite many female pilots, so agents scouted inventors and companies across the planet, and then a hundred of those people were brought together in person. That combination produced rapid prototypes, and a similar concept later flew on the Artemis mission. His point: a fraction of that achievement lived in the software, and most of it lived in humans working together.He closes with a framework for collaboration in four parts: human to human, human to agent, agent to agent, and agent back to human. He argues that anyone who understands where they sit in that flow will succeed, because control comes from understanding the system rather than from the tool itself. He also says nobody has solved the company brain everyone is chasing, and suggests quantum may finally let computing stop being binary the way humans already are. Along the way he cites a claim from one of the founders of Claude Code that only about five thousand people in the world truly understand how LLMs and agents operate, which he uses to argue that most leaders need to spend more time actually using the tools.
Hackers do not need your password. That is the warning at the center of this episode, and the story that proves it came from one of the world's leading AI companies. Last week, OpenAI found a recording device hidden behind a TV in one of its executive rooms, reportedly concealed in toilet paper and plugged into the television. An employee was arrested and charged. No network was breached. No credential was stolen. Someone simply walked in and left something behind. This week on Machine Dreams, Sam Adebayo and Leah Stern sit down with Shmulik Yehezkel, Col. (Res.), founder and CEO of Cygon (https://www.cygon.io/). Shmulik spent more than 25 years securing defense, intelligence, and government environments in Israel. He headed the Mossad School for Cyber and Operational Technology and served as national CISO for Israel's defense industrial base before founding Cygon in 2024. His argument is that this was not a failure of cybersecurity spending. It was a structural blind spot. We secure the network, the cloud, and the endpoint, but the attacker who comes close to the facility almost always succeeds, because that proximity attack surface sits outside the network where traditional defenses cannot see it. Inside any building there is unmanaged IoT, unmanaged IT, and people, which makes it, in his words, like a kid in a candy shop. He notes the device reportedly cost about $14, and points to commercial off-the-shelf tools available online for around $50. Then there is Flipper One, an AI autonomous hacking tool that sells for roughly $350 and can open doors, jam networks, and launch cellular attacks. His line: you do not need to be a cyber hacker anymore, you just need $300 and an internet connection. He also shares that more than 55 percent of the incidents Cygon identifies start with an insider, including a SOC team member who used a Flipper Zero to open the CIO's office door at night. He closes by arguing that AI infrastructure has become critical infrastructure and should be treated like a new embassy, secured not only remotely but against boots on the ground. The attacker lives in the seam between the physical and digital domains, so the answer is holistic security rather than more of the same spending. Along the way he shares incidents that go well beyond data, including a fired employee who shut down a five-star hotel's electricity for 48 hours from his own home, and a Bluetooth spam attack that sent bomb threats to passengers and forced the Israeli Air Force to scramble two F-15s. If you are responsible for securing an organization, this conversation is a grounded place to start.
AI agents can answer questions, handle repetitive work, and complete tasks across applications. But the real question is not what they can do in a demo. It is what happens when one makes a mistake, gets access to information it should not have, or simply cannot do the task.This week on Machine Dreams, Sam Adebayo and Leah Stern sit down with Nate Herk, founder and CEO of Uppit AI (https://www.uppitai.com/), the education platform behind the AI Automation Society. Nate left Goldman Sachs as a business intelligence analyst to build an AI consultancy, scaled it to $100,000 a month, and exited in December 2025. He now runs a YouTube channel with more than a million subscribers, and he is the Amazon best-selling author of Becoming AI Native.So what does it really take to succeed? Nate's answer starts before any tool is chosen. A large share of the companies that call him are not buying AI, they are buying relief. They need to tell their board, their employees, or their stakeholders that they have an AI strategy. His advice is to find the actual problem first. He also draws a clear line on automation. Not everything that can be automated should be, and he uses his own community manager as the example, a role that could be 95 percent automated but is one of the reasons members stay. His framing for agents is simple and practical. You onboard an agent the way you onboard a person, you give it context, you keep the guardrails on, and you let it earn more autonomy over time. You would not hand an intern a credit card to a real bank account and tell them not to use it.That is where the real advantage sits, and it is not in the tools. Anyone can spend a weekend and emerge with a demo that looks like two years of experience, so what separates people is what cannot be faked. Governance, meaning who owns the system and what happens when it fails. Evals, proving it works over a thousand runs and that its failures are safe. And stakeholder communication, because the operator whose workflow is changing is often the last person told. He also walks through the small stack of tools he actually uses to run his business.
When you ask an AI a question about your company's data, it looks something up before it answers. Most of the time that works. But often it joins the wrong tables and returns an answer that looks completely reasonable and is completely false, and nothing catches it. There is no error message. This week on Machine Dreams, Sam Adebayo and Leah Stern sit down with Weimo Liu, CEO and co-founder of PuppyGraph https://puppygraph.com​ Weimo previously worked on query systems at Google and spent time at TigerGraph, and he now helps write the international standard for how these systems get queried.PuppyGraph's approach is to check the question before the AI runs it. The product was designed for agents as the user, not humans, and a semantic layer sits over the enterprise data so the AI only needs to know that a teacher teaches a student, not which table holds which or how to join them. The engine handles the deterministic mapping and guarantees accuracy, so the AI reasons and something outside the model verifies. Weimo also covers the Sonnet 4.6 inflection, where early work with customers including AMD became reliable after proper context engineering, how Coinbase runs long-running compliance agents in natural language without writing a single query, and why chaotic enterprise data needs an interactive human-in-the-loop agent rather than automation.He closes with the warning that stayed with us. One AI agent writes a financial report, another AI agent reviews it, and no human looks at how it was generated, so nobody can know whether the answer is right. He contrasts this with code, where test cases can cover the gaps, and names the human factor plainly: with a tool this powerful, nobody wants to check every detail.
Is AI really conscious? This week on Machine Dreams, Sam Adebayo and Leah Stern sit down with Andy Thurai, founder and principal analyst at The Field CTO https://thefieldcto.com/ an independent technology advisory firm helping executives turn complex technology into clear business strategy. Andy brings more than 25 years of enterprise technology experience, with senior roles at IBM, Intel, and Oracle, and now advises business and technology leaders on AI, observability, and modernization.Andy's answer to the consciousness question is careful and grounded. Machines today have simulated awareness, not self-awareness. His sharpest illustration: throw hot water on a dog and it reacts immediately because it feels pain. Throw hot water on a robot and it only reacts if someone trained it to. That gap is the difference between real consciousness and simulation, and it is also where AI fails, when unknown unknowns arrive. He then offers his most provocative claim: the rush to prove AI is conscious may not be about philosophy at all. If a model can be shown to act on its own, its creators can step back from liability when it causes harm. He also draws a firm line on ethics. When an AI refuses an unethical prompt, it is not consulting a conscience. It is calculating which answer yields a higher reward. It is, in his words, an advanced calculator doing statistical ethics.He closes with practical advice for enterprise leaders. Work backwards from the problem instead of choosing a model and hunting for uses. Treat the LLM as a reasoning engine, and put safety, compliance, and guardrails outside the model in a deterministic control plane. Do not ask a probabilistic system to police itself, because it is architecturally impossible. His summary line: don't trust, always verify.If you are trying to separate the real AI risks from the noise, this conversation is a grounded place to start.
loading
Comments