The AI Riff

AI’s biggest bottleneck may not be smarter models—it may be power, water, and the massive data centers quietly reshaping entire communities. In this episode, the team digs into the explosive race to build AI infrastructure, unpacking why hundreds of billions are flowing into data centers, why local residents are pushing back, and whether the industry is headed toward a construction boom—or a painful overbuild.

From the surprising reality that AI usage consumes more resources than training, to the debate over tax incentives, energy grids, climate impact, and the hidden economics driving the AI arms race, this conversation separates headlines from reality. If you’re a tech leader, operator, investor, or simply trying to understand what’s powering AI behind the scenes, this episode explores the tradeoffs, risks, and unanswered questions most people are missing right now.

Agentic AI is supposed to be the next big leap—but what happens when the real problem isn’t the AI, it’s everything around it? This episode unpacks a revealing industry study showing that companies aren’t struggling because AI is too dumb; they’re struggling because deploying, governing, and maintaining AI at scale is far harder than expected.

The conversation explores the hidden costs of AI orchestration, why hallucinations become exponentially more dangerous inside complex workflows, the risks of vendor lock-in, and the surprising new careers emerging from the AI boom. From context management and governance to data center growth and the future of agentic systems, the hosts challenge both the hype and the fear—while revealing what most organizations are missing right now. If you’re a founder, operator, technologist, or AI leader, this is a reality check you can’t afford to ignore.

AI's biggest players are racing toward historic valuations—but who ultimately pays the bill?

As OpenAI, Anthropic, and SpaceX move closer to blockbuster IPOs, the conversation shifts from innovation to economics. Will public market pressure force AI companies to raise prices, monetize through ads, or fundamentally change how businesses use AI?

The team explores the growing uncertainty around token costs, whether startups can survive if AI becomes more expensive, and why today's low barriers to building AI-powered products may not last forever.

From mortgage tech and enterprise adoption to open-source competition, data centers, and the future of AI infrastructure, this episode tackles one of the biggest questions facing founders, operators, and technology leaders right now: what happens when AI has to start paying for itself?

What happens when AI stops waiting for instructions and starts creating its own agents on the fly?

In Episode 51 of The AI Riff, the team explores the emerging concept of self-evolving AI agent factories—systems capable of spawning specialized AI agents to solve problems they've never encountered before.

The conversation dives into the promise and peril of autonomous AI, from mortgage lending use cases and workflow automation to the hidden risks of token consumption, model drift, security vulnerabilities, and trust.

The discussion expands into a broader debate about where AI is already exceeding expectations, where human oversight remains essential, and which professions and skills may prove more resistant to automation than most people realize.

If you're curious about the future of AI agents, AGI, automation, governance, and the real-world limits of autonomous systems, this episode delivers a thoughtful, practical perspective without the hype.

AI isn’t coming for the mortgage industry “someday” — it’s already knocking on the door. In this episode, the crew unpacks Andreessen Horowitz’s bold idea of moving from “systems of record” to “systems of intelligence,” and what that could mean for lenders drowning in legacy tech, APIs, CRMs, and disconnected workflows. From AI-powered loan officer assistants that can answer calls in real time to the looming cost explosion of AI tokens and API access, the conversation gets surprisingly practical — and controversial.

They debate whether AI will truly become agentic enough to replace human expertise, or if the real future is smarter orchestration layered on top of existing systems. Along the way: compliance headaches, AI licensing, the future of loan officers, and why the mortgage industry may be closer to disruption than most people realize.

a16z article: https://open.substack.com/pub/a16z/p/from-system-of-record-to-system-of

What happens when AI stops acting like a chatbot—and starts understanding how your business actually works?

In this episode, the crew dives into the emerging world of “context models”: AI systems trained not just on the internet, but on the messy, contradictory, highly specific realities of how companies operate day to day.

From mortgage compliance and workflow chaos to hallucinations, hidden operational risk, and the myth of every lender’s “secret sauce,” this conversation gets into the uncomfortable gap between how businesses *say* they work and how they *really* function.

The debate gets surprisingly sharp: should AI preserve a company’s uniqueness, or force industries toward standardization and efficiency?

And if every organization thinks it’s different, can a true industry-wide AI model ever exist? For operators, tech leaders, and anyone watching AI reshape enterprise software in real time, this episode surfaces the questions most companies still aren’t asking.

References: Celonis Context Model: https://www.celonis.com/news/press/celonis-launches-the-context-model-to-eliminate-enterprise-ais-operational-blind-spots-agrees-to-acquire-ai-decision-intelligence-leader-ikigai-labs

What happens when AI starts making decisions your business can’t fully explain—and your customers, regulators, or employees are the ones who find the mistakes first? In Episode 47 of The AI Riff, Kyle, Mike, and Mario dive into one of the most urgent—and massively underbuilt—frontiers in enterprise AI: AI observability, agent evaluation, and red teaming.

From hallucinations and prompt drift to shadow AI inside mortgage lending, banking, healthcare, and other regulated industries, the crew explores why today’s AI systems still feel “squidgy around the edges”—and why the companies solving trust, guardrails, auditability, and AI governance may become the next billion-dollar acquisition targets. They also tackle a bigger question most leaders aren’t asking yet: when AI starts influencing culture, compliance, and strategic decisions… who’s actually watching the algorithm? And how long before the first public AI failure makes headlines? Their answer: maybe six months.

What happens when AI stops being a chatbot—and starts becoming your front line? In this episode, the hosts unpack one of the boldest predictions from CB Insights: multimodal AI agents that can talk, text, email, and even appear on video may soon transform customer service. But here’s the real question—will they actually improve the customer experience, or just become a more expensive version of the old phone tree?

From token costs and latency issues to ROI, data training, human handoffs, and the “sameness tax” of off-the-shelf AI, this conversation digs into what most companies are missing right now. The hosts challenge the assumption that AI automatically reduces headcount, explore why hybrid human + AI models may win, and make a bold prediction: within 2.5 years, customers may be escalated… from humans to AI. For founders, operators, and customer experience leaders, this is the AI adoption conversation happening in real time.

A $1.8B “AI startup” built in two months—genius growth story or something far more dangerous? This episode pulls back the curtain on how AI can manufacture credibility at scale, from fake doctors to synthetic social proof, and why the real risk isn’t just bad actors—it’s how easy the tools have made it.

The conversation digs into a deeper tension: AI as both accelerator and amplifier—of innovation and fraud. From compliance blind spots in regulated industries to Meta’s evolving role in policing AI-generated content, the hosts explore what happens when speed outpaces oversight. If you’re a builder, operator, or leader navigating AI right now, this is the uncomfortable reality check most people are still ignoring.

A new AI model just found software vulnerabilities that have been hiding for 26 years—and its creators are too cautious to release it. That’s the tension at the heart of this episode: breakthrough or carefully staged hype?

We unpack Anthropic’s mysterious “Mythos” model, the secretive Glass Wing rollout, and what it really means when AI can chain together exploits like a seasoned penetration tester. Is this a turning point for cybersecurity—or a clever enterprise play ahead of an IPO? Along the way, we confront the uncomfortable reality: AI-powered hacking isn’t coming… it’s already here. For founders, operators, and anyone building in tech, this is the conversation you can’t afford to ignore right now.