Episode 4 of Multipliers: AI Crossed the Threshold

AI is not replacing expertise. It is revealing who has it.

The professionals who have spent years building real judgment, real methodology, and real domain knowledge are discovering something unexpected: AI does not compete with them. It amplifies them. The ones who have been going through the motions are discovering something else entirely.

In this episode of the Multipliers podcast, the team unpacks what it actually looks like when a subject matter expert learns to orchestrate AI, and why the conductor metaphor may be the most useful mental model for the next phase of knowledge work.

• You might also enjoy: Spencer’s article that kicked off this conversation: AI Has Crossed a Threshold

• Related: The Multiplier Framework Workshop #3: AI Skills, where Spencer walks through how to encode your expertise into reusable AI skills

Episode 4 of Multipliers: AI Crossed the Threshold

This episode is a conversation between Spencer Burton, Michael Belasco, and Sam Carlson, the three partners behind Adventures in CRE and CRE Agents.

Like Episode 3 on the coming AI divide, this is an internal conversation rather than a guest interview. It picks up where that episode left off, moving from the macro question of who will be affected to the practical question of what it looks like when someone leans in. Michael has been building something remarkable over the past month, and the conversation uses his experience as a lens into a bigger idea: the shift from using AI to orchestrating it.

Why This Episode, Why Now

Something shifted in the weeks between Episode 3 and this recording.

Michael came home from a working session in Miami with Spencer and Gert (CRE Agents’ CTO) and entered what he describes as a frenetic creative stretch. He started building a bespoke agentic AI platform for his outdoor hospitality business, not by licensing software but by orchestrating a team of AI agents to handle operations, pricing, reviews, and market intelligence. The kind of system that would have required a full engineering team and six figures of development budget 18 months ago.

Spencer places this in a broader context. Around early December 2025, AI coding agents crossed a threshold. Andrej Karpathy, the former head of AI at Tesla and OpenAI co-founder, posted that programming had become “unrecognizable.” Before December, AI coding tools were useful novelties. After December, they worked. For non-technical domain experts like Michael, this unlocked something that the team had been predicting for a year but that only recently became real: the ability to build sophisticated, custom systems without writing a line of code yourself.

At the same time, Spencer had been tracking Howard Marks’ evolving view on AI. In December, Marks asked whether AI was a bubble. By late February, his position had shifted significantly. He wrote that he believed we were more likely underestimating than overestimating AI’s potential, and expressed concern that the speed of dislocation could outpace the workforce’s ability to adapt.

These threads, Michael’s hands-on experience, the coding threshold, Marks’ shift, converged into a recording that is less about whether AI matters and more about what it looks like when someone with real expertise starts conducting.

Episode Highlights

Here are the themes that stood out.

1. The Coding Threshold

The episode opens with context that most non-technical professionals have not fully internalized: something fundamental changed in AI coding capability around December 2025.

Spencer walks through the timeline. Before December, AI could generate code, but it was unreliable. For experienced developers, it was a helpful assistant. For non-developers, it was a novelty that produced buggy outputs you could not fix. Michael references an early ChatGPT experience where he let a coding project run overnight and came back to a mess of bugs. It was cool. It did not work.

Then it started working. Not incrementally. Categorically. Spencer cites Karpathy’s observation that programming had become unrecognizable in a matter of weeks. The implication for knowledge workers outside of tech is enormous: the barrier between having an idea and building a functional system collapsed. You no longer need to be a developer. You need to know what you want built and why.

That distinction, knowing what versus knowing how to code, is what makes subject matter expertise the new bottleneck. Michael has 20 years of outdoor hospitality experience. He knows what operational intelligence should look like, what pricing decisions require, and what guest experience data matters. AI handles the implementation. He provides the judgment.

2. The Conductor Metaphor

Michael describes his new working rhythm as feeling like a conductor of an orchestra. The metaphor is not casual. It captures something specific about how expert-led AI orchestration differs from basic AI usage.

A conductor is not the best violinist in the room. They are not the best cellist or oboist. What they bring is the ability to hear the whole piece, to know when something is off, to shape the output of many specialists into a coherent performance. That is exactly what Michael is doing with his AI agents. He is not writing code. He is not designing databases. He is orchestrating a team of AI capabilities around his domain knowledge, and the result is a system that reflects his expertise at every layer.

Spencer adds the Seinfeld reference (the “Maestro” episode, for the uninitiated) but the underlying point is serious. Sam reinforces it with a distinction he picked up from the All In podcast’s discussion of Anthropic hiring software engineers: the engineers who are thriving are not the ones who just write code. They are the ones who see product development and code simultaneously. They are orchestrators. The builders who lack that higher-order vision are the ones struggling.

For CRE professionals, the parallel is direct. The analyst who just runs numbers is more exposed than the one who understands why those numbers matter and can orchestrate AI to produce the analysis.

3. The Tsunami on the Beach

Spencer offers a visual that stays with you. We are all sitting on the beach. The weather is nice. And then the water starts going out. Some people look at the extra sand and think they have more room to play. Others recognize what a receding waterline means: a wave is coming.

This is not presented as doom and gloom, though Spencer acknowledges the gravity of it. His view is that there is a higher probability that fewer people will be needed in knowledge-work careers than more, though he leaves room for the opposite outcome. Either way, the divide is forming between the 5 percent who are already orchestrating AI around their expertise and the 95 percent who are still using it as a glorified search engine.

He ties this to Howard Marks’ February memo, where Marks expressed concern that the pace of AI-driven dislocation could outrun society’s ability to adapt. Spencer sees the same pattern forming in real estate. At three recent industry conferences, almost no one was seriously discussing the second-order effects of AI on markets, submarkets, or headcount.

Michael’s response is more personal. He describes his energy as frenetic. The possibilities are so vast and moving so fast that his mind races in every direction. But he also offers a utopian counterpoint: what if this creates a world of a million tiny entrepreneurs? What if AI enables individuals to break off from large organizations and serve their own communities with bespoke services, running micro businesses powered by AI workforces that would have required 50 employees five years ago?

It is an optimistic vision, and Michael acknowledges it is the “100 percent optimistic” version. But it is grounded in what he is actually experiencing. If he can build a full operational platform for his business using AI, so can the accountant down the street, the landscaper, the local property manager.

4. Bespoke Beats SaaS

One of the most provocative threads in the episode is what Michael’s platform means for commercial software.

Spencer makes the point explicitly: most of what Michael has built already existed as licensed software. Review management tools, dynamic pricing platforms, market intelligence dashboards. They all exist. But they are expensive relative to what Michael pays for AI. They are disparate, requiring logins to 10 different systems. They are built for the lowest common denominator, not bespoke to Michael’s business. And they were designed by someone else.

Michael’s system consolidates all of that into one platform, built around his specific workflows, his specific data sources, his specific decision-making process. Dynamic pricing is the clearest example. The concept is not new, but implementing it used to require either a massive organization or an expensive third-party tool that offered static rules. Michael now has real-time market data flowing into a system that makes pricing recommendations every morning, tailored to his properties, his competitive set, his occupancy patterns.

Even the “bells and whistles” illustrate the shift. Review monitoring across seven platforms used to require a team member spending 30 minutes each morning logging into each site. Now the system aggregates reviews overnight, generates proposed responses, and presents them for approval. The human provides the final judgment. AI handles the collection, analysis, and drafting.

Spencer extends this to a broader observation: if one operator in outdoor hospitality can build this, what happens to the SaaS companies that sell these tools piecemeal? And what happens to Michael’s competitors who are still running their businesses the old way?

5. Chat Is a Prototype

Spencer shares a quote from Gert, his CTO at CRE Agents, that reframes how most people interact with AI: chat is a prototype for this technology.

The observation lands because it challenges the default assumption. Most professionals equate “using AI” with opening ChatGPT or Claude and typing a prompt. Gert’s point is that this interface, the back-and-forth text conversation, is not the destination. It is an early-stage interface that will look primitive in hindsight.

Michael’s experience illustrates the difference. His platform is agentic. AI agents monitor data, generate recommendations, and surface insights without being prompted. When he built a chat feature into the platform, it was not the primary interface. It was a secondary tool he added so he could ask the system why it made a particular recommendation. Chat went from being the doorway to AI to being a feature within a larger system.

Spencer connects this to the broader frustration he sees across the industry. People move from ChatGPT to Claude as if switching chat providers will fundamentally change the experience. It will not. The leap is from chat-based interaction to orchestrated, agentic systems where AI operates continuously and the human provides oversight, judgment, and direction. That is what Michael has built. That is where the industry is heading.

6. AI Skills as Intellectual Property

The episode’s closing thread may be its most important: the idea that your expertise, encoded into AI, becomes intellectual property worth protecting.

Spencer makes the case directly. Every experienced CRE professional carries methodology in their head. How they underwrite property tax. How they structure a debt module. How they evaluate a market. That methodology has always been valuable, but it was locked inside the individual. It could only be deployed one task at a time, at the speed of one person.

AI skills change that equation. When you encode your methodology into a reusable AI skill, you create something that can be deployed across hundreds of tasks simultaneously, with consistent quality. Spencer references the AI Skills workshop the team ran the day before recording, where they walked through exactly this process.

But there is a tension. If you upload your methodology to a general-purpose AI platform, you may be giving away your competitive advantage. Michael’s approach is different: he encoded his expertise into a proprietary system that only his company uses. That is IP protection through architecture, not through secrecy.

Sam brings a complementary angle. His entire workflow is built around frameworks: repeatable, structured methodologies for everything from writing ad copy to building sales engines. He describes listening to a colleague walk through a lead generation system and immediately diagramming it into a framework he could encode. The power, Sam says, is duplication. Once your methodology exists as an AI skill, your team can use it without you. You are no longer the bottleneck.

Spencer closes with a reference to a Y Combinator partner’s recent post: one of the biggest opportunities right now is finding existing businesses operating in analog fashion and bringing AI into them. Encode 30 years of a landscaper’s knowledge into a system that doubles their capacity with the same overhead. That is the maestro effect applied beyond the office.

The Bigger Idea

There is a line in this episode that ties everything together.

Spencer references a conversation with Gert where they land on a principle: chat-based AI is a prototype. What comes next is orchestration, agentic systems where AI operates on your behalf, governed by your expertise, and you provide the judgment layer on top.

Michael is living that principle already. He is not prompting AI one question at a time. He is running a bespoke platform that monitors, analyzes, and recommends, with him as the conductor deciding what gets played and what gets cut. And his competitors, who are not doing this, are falling behind in ways they may not yet recognize.

But the episode is not just about Michael’s platform. It is about a mental model shift that applies to every knowledge worker. The 95 percent of CRE professionals who are still in chat mode are not behind because they lack intelligence or ambition. They are behind because they have not yet experienced what the technology can do when paired with real expertise. They have not felt the moment Michael describes, where all of a sudden you have an orchestra and you realize you know how to conduct.

The advice from all three hosts converges on action. Sam points to AI.Edge as a starting point for building foundational AI skills and encourages people to participate in the build competition, not because the builds are perfect, but because building is how you learn. Michael emphasizes community: the breakthroughs he has had came not from sitting alone with AI but from working sessions with Spencer and Gert where human expertise sharpened the AI’s output. Spencer floats the idea of a CRE hackathon, a hands-on event where practitioners build together.

The maestro effect is not about replacing anyone in the orchestra. It is about having something worth conducting. If you have spent years building expertise, judgment, and methodology, AI is the instrument that lets you play it at scale. The question is whether you will pick up the baton.

---

Frequently Asked Questions about Episode 4 of Multipliers: The Maestro Effect