By Don Finley | The Human Code Podcast
Everyone wants to know if we’re in an AI bubble. I think that’s the wrong question for most business leaders.
The spending looks shaky. Hyperscalers are pouring billions into data centers. Chipmakers invest in AI labs, and those labs turn around and buy the chips. Returns on that spending are nowhere near what investors were promised. By most measures this looks like a bubble, and it has looked like one for more than a year.
Here’s the uncomfortable truth: if it pops, AI doesn’t go away. And if it doesn’t pop, the machines still won’t run your business for you. Either way, the organizations that win will be the ones that pair AI with human judgment.
Why the AI Bubble Question Matters Now
I recently sat down with Andre Chulo, a senior at the Wharton School at the University of Pennsylvania who studies finance and AI, on The Human Code podcast. He’s about to enter a job market that AI is actively reshaping, and he’s building a startup with fellow students. That puts him in a useful spot: he has to bet his career on where this technology is really going, not where the press releases say it’s going.
His read matches what I see with our clients at FINdustries. The hype and the spending are real, and so is the uncertainty. The value, though, doesn’t come from the model by itself. It comes from the people directing it.
Every Bubble Leaves Something Behind
We’ve been here before. In the dot-com era, Cisco became the poster child for overbuilding. We ended up with miles of unused “dark fiber” that sat idle for years. Then the bubble burst.
The internet didn’t stop. The fiber got used. And the companies that came out on top, like Amazon, weren’t the ones selling picks and shovels at the peak. They were the ones whose business models didn’t depend on the hype holding up.
Andre put it well:
“Even if we see some sort of market correction, some sort of bubble collapse, we’re not going back. AI is only gonna continue. Just like when the dot-com bubble burst, we didn’t see a stop in the internet.”
The lesson for anyone running a business is simple: build something that doesn’t depend on any one AI company surviving the next 24 months.
Andre sorts AI businesses into two camps. The first builds right next to the frontier labs, reselling or lightly wrapping their models. That’s profitable in the short run, but “if we see some sort of market correction, you’re also gonna be the first to go.” The second camp solves the knock-on effects AI creates in the real world. His example is generative engine optimization (GEO), which is SEO for AI answers: how does your brand show up when someone asks ChatGPT for a recommendation? Those problems exist no matter whose model wins.
I’d add a third category that matters most for established businesses: putting today’s models to work inside your own workflows. Even if frontier models stopped improving tomorrow, most companies have barely started capturing the value that’s already available. Large enterprises have the budgets and teams to push agents through that change-management hump. Small and mid-sized businesses mostly don’t yet. That gap is where the real opportunity is.
Why Human in the Loop AI Beats Full Automation
Here’s something I’ve watched play out over and over, with clients and inside our own team.
People usually fall into two camps with AI. Some accept whatever it produces. Others stay skeptical of everything. Both extremes are wrong, but the skeptics consistently get better results. They ask better questions, push back on weak answers, and steer the system toward what actually matters.
We work with financial services firms, and my background is technology. When I ask our AI system for a financial analysis, I get a decent answer. When the CEO of one of our clients uses the same system, he gets insight I can only touch the surface of. The difference isn’t the model. It’s his expertise finding a path through the intelligence.
Andre compared it to a calculator:
“You have to learn how to do math to use a calculator. You can’t just have a calculator.”
That’s why there’s so much AI slop out there. People generate without direction. The model doesn’t know what makes your business special unless someone who knows teaches it. Your expertise is the input that turns generic output into a competitive advantage.
That’s the principle behind how we build Sofia, our AI platform: human in the loop, agent in the work. The agents do the repetitive work. Your people give direction, check the output, and make the calls that need judgment. Every action stays visible and accountable, because trust has to be designed in. You can’t bolt it on later.
The Skills That Survive Any AI Future
So how should a professional, whether they’re graduating in eight months like Andre or running a 50-person firm, prepare?
Andre’s answer surprised me. He’s deeply immersed in AI tools, but he takes seriously a school of thought that says early-career people should focus less on the tools and more on fundamentals: structuring an argument, doing real research, thinking critically. The tools will look completely different in five years. The ability to think clearly won’t.
“It’s about what you can get out of the technology, not about what the technology can get out of you.”
He also pointed to relationships as the most durable asset of all: “If you’re able to be a warm face, you’re able to establish a strong network, that’s something that an AI is not gonna be able to do for you.”
I’d frame the question this way: where are you a user of AI, and where are you a builder with it? I’m happy to let Netflix’s algorithm pick my next movie. But how our business reaches customers, measures performance, and makes decisions? That’s where we stay in the loop. Those are the areas where understanding your customers deeply is your moat, and AI can help with that but isn’t ready to drive it.
Beware Lock-In While the Giants Fight It Out
One more risk deserves attention. The frontier labs are pushing for regulation, and some of that is genuine concern about where AI is headed. But as Andre pointed out, some of it looks like an attempt to build a regulatory moat against open-weight competitors.
For businesses, that makes vendor dependence a real strategic risk. If a single provider decides what you can and can’t do with AI, your operations are only as stable as their policies, pricing, and balance sheet. A healthy open-source and open-weight ecosystem gives companies options. Andre summed it up: no single group “should kinda determine our fate, even if they have our best interests in mind.”
That’s why we built Sofia to be model-agnostic, so our clients own their workflows and their context no matter which lab is ahead next quarter.
How to Build a Bubble-Proof AI Strategy
Whether the AI bubble bursts, deflates slowly, or keeps inflating, these moves hold up:
- Start with a process, not a vision. Visionaries matter, but the fastest wins come from a specific workflow: “This costs us $100 every time, and here’s where the human gets involved.” If AI can do it for $20, the value is obvious on day one.
- Put your experts in the loop. Your best people’s judgment is what makes AI output worth using. Design workflows so the agent does the work and the human directs and reviews it.
- Avoid single-vendor dependency. Pick platforms that let you switch models without rebuilding everything.
- Invest in human skills alongside AI skills. Critical thinking, clear communication, and relationships go up in value as generation gets cheaper.
- Teach the AI what makes you different. Generic prompts produce generic results. Capture your business context so the system works like you do.
Ask yourself: if your AI vendor raised prices 3x or shut down next year, what would break? If the answer is “everything,” that’s your first project.
The Human Code Is the Real Moat
The AI bubble debate will keep going, and nobody, including me, knows exactly how it ends. What I do know is that the technology is a tool, and what you get out of it depends on the human directing it.
The companies that come through this era stronger won’t be the ones that spent the most on AI. They’ll be the ones that combined their people’s judgment with agents doing the work. That’s the human code.
Want to put AI to work in your business without betting it on the hype cycle? Talk to FINdustries about how Sofia brings AI agents into your workflows, with your people in control.
🎧 Listen to the full conversation with Andre Chulo on The Human Code podcast.
Related Reading
- AI Replaces Intelligence, Not Judgment: Why human judgment becomes more valuable as AI gets smarter.
- AI Valuations vs. AI Infrastructure Returns: A closer look at the gap between AI spending and AI returns.
- The AI Memory Trap: Why Your Frontier Model Vendor Is Becoming Your Most Dangerous Dependency: The hidden cost of AI vendor lock-in.
- Why Your Business’s AI Strategy Should Start with the Boring Stuff: Where practical AI ROI really comes from.
- AI Adoption Is a People Problem (And That’s the First Thing You Need to Accept): Why AI adoption succeeds or fails on people, not tools.
Don Finley is the founder of FINdustries and host of The Human Code podcast. He helps financial services firms and growing businesses put AI agents to work, with people in control. Subscribe on Apple Podcasts, Spotify, or wherever you listen.
