Welcome back to the show, everyone. It is a beautiful, crisp morning here in Jerusalem. The sun is hitting the stone walls of the Old City just right, and I am sitting across from my brother, Herman, who I suspect has spent the better part of the morning buried in proprietary market research, three different newsletters, and drinking way too much of that high-altitude espresso he loves.
Herman Poppleberry here, and you are absolutely right on all counts, Corn. I have been neck-deep in the latest quarterly trend reports from three different firms. It is a sickness, really, an obsession with trying to see around the corner. But it is a fascinating time to be watching the world. We are living through a period where the "future" is arriving in weekly installments, and honestly, it is getting hard for even the professionals to keep up.
Well, your timing is perfect because Daniel’s prompt today is right up your alley. He wants to talk about the rapid rise of artificial intelligence and how it is transforming basically every industry we can think of. But specifically, he wants us to look at it through the lens of the Gartner Hype Cycle and other frameworks that chart how technology actually evolves from a shiny new toy into something we actually use to get things done.
That is such a great angle. People often look at technology as a straight line, like it just gets better and better at a constant, predictable rate. We have this mental model of a ladder where we just keep climbing. But the reality of human adoption is much more of a roller coaster. It is emotional, it is financial, it is sociological, and it is incredibly messy. We do not just adopt tools; we fall in love with them, we get angry at them when they fail us, and eventually, we just forget they are there because they become part of the furniture.
It really is a psychological journey as much as a technical one. And Daniel specifically asked where we would plot artificial intelligence as a whole right now—today, in February twenty twenty-six—and then more specifically, where agentic A-I sits on that map. So, Herman, for the people who might have heard the term but do not live and breathe market research like you do, can you give us the quick and dirty on what the Gartner Hype Cycle actually is? Why do we use this specific shape to describe progress?
Right, so the Gartner Hype Cycle is a graphical representation that the research firm Gartner has used since the mid-nineties to track the maturity and adoption of new technologies. It is based on the idea that every new innovation goes through a period of over-enthusiasm followed by a period of disappointment. It has five distinct phases. First, you have the Innovation Trigger. That is the breakthrough, the first public demo, the "holy cow" moment that gets the tech press screaming. Then you climb the Peak of Inflated Expectations. This is the "frenzy" phase. This is where the hype is out of control. Every startup is claiming they will change the world, every C-E-O is being asked by their board what their strategy is, and everyone is terrified of being left behind.
That sounds exactly like twenty twenty-three and twenty twenty-four for A-I. I remember every single email in my inbox had "A-I" in the subject line.
Exactly. It was a gold rush. But then comes the scary part: the Trough of Disillusionment. This is where the tech hits the cold, hard wall of reality. Implementation is harder than people thought. The return on investment—the R-O-I—is not immediate. The "magic" turns out to have edge cases and bugs. This is when the press starts writing articles about how the whole thing was a bubble and a waste of money. But—and this is the important part—if the tech is actually fundamentally sound, it enters the Slope of Enlightenment. That is when the "tourists" leave and the "builders" stay. We start to understand how to actually use it. It is less about the hype and more about the hard, boring work of integration. Finally, you reach the Plateau of Productivity. The tech becomes "boring," which is actually the ultimate goal. It is just a tool we use every day without thinking about it, like electricity or the internet.
I love that "boring" is the final stage. It is like when a magic trick becomes a utility. It loses its wonder but gains its value. So, let us tackle the big question Daniel posed. If you had to put a single dot on that map for artificial intelligence as a whole, right now, in February twenty twenty-six, where are we?
Man, that is a tough one because A-I is not a single thing anymore. It is like asking where "computing" or "the internet" is on the hype cycle. It has become too broad. But if we are talking about the broad category of large language models and general generative A-I—the stuff that blew everyone's minds a few years ago—I think we are currently sliding down from the Peak of Inflated Expectations and heading straight into the Trough of Disillusionment.
Really? You think we are entering the trough? That feels a bit pessimistic given how much we are still seeing it everywhere. I mean, I can't open a spreadsheet without an A-I offering to summarize it for me.
It is not necessarily pessimistic; it is just the natural cycle of things. Think about the sentiment shift we have seen over the last twelve months. In twenty twenty-three and twenty twenty-four, the expectation was that A-I would replace every white-collar job by next Tuesday. Companies were throwing hundreds of millions of dollars at anything with a dot A-I domain name. Now, here in early twenty twenty-six, the boardrooms are asking much harder questions. They are saying, "Okay, we spent fifty million dollars on this integration, we bought ten thousand enterprise licenses, where are the actual savings? Why is our productivity only up by two percent instead of forty percent?" We are seeing the limitations. We are seeing the massive cost of compute. We are seeing the legal challenges over training data that are finally hitting the courts. The "magic" is wearing off, and the "work" is beginning. That transition from magic to work is exactly what the Trough of Disillusionment feels like. It is the "hangover" phase.
That is a fair point. I think there is a massive gap between what the models can do in a vacuum—like in a clean demo on a stage—and what they can do inside a complex, messy corporate database that hasn't been cleaned since nineteen ninety-eight. I have seen so many companies realize that their internal data is such a disaster that no amount of A-I can make sense of it without a year of manual cleaning first.
Precisely. The "A-I" part is the easy part now; the "data engineering" part is the hard reality that leads to the trough. We are realizing that A-I is not a magic wand; it is a very sophisticated engine that requires very high-quality fuel. But here is the interesting twist, and this gets to the second part of Daniel's question. While general generative A-I might be dipping into that trough, the specific sub-category he mentioned—agentic A-I—is currently screaming up the side of the Peak of Inflated Expectations. It is having its own mini-cycle within the larger one.
Okay, let us pause there and really dig into that. Agentic A-I. We have talked about this a little bit in passing over the last few episodes, but for this deep dive, let us define it clearly. What makes an A-I "agentic" versus just being a very smart, very fast chatbot?
The simplest way to think about it is the difference between "thinking" and "doing." A standard chatbot, like the early versions of ChatGPT or Claude, is reactive. You give it a prompt, it gives you an answer based on its training data. It is a sophisticated autocomplete. It is a "stateless" interaction—you talk, it responds, the end. Agentic A-I, on the other hand, is proactive, stateful, and goal-oriented. You do not give it a prompt for a single answer; you give it a high-level goal and the authority to act.
So, instead of asking, "Give me a list of hotels in London," I say, "I need to go to London for a three-day business trip next month. My budget is two thousand dollars, I need to be near the Shard, and I prefer Marriott properties. Go plan it and book everything."
Exactly. An agent can reason through a multi-step process. It has what we call a "reasoning loop." It can use tools. It can browse the web to check current prices, use an A-P-I to check your personal calendar for conflicts, send an email to a colleague to confirm a meeting time, wait for a response, and then adjust its plan based on that response. It observes its environment, decides on an action, executes that action, and then evaluates the result. If the result wasn't what it expected, it tries a different path. That "loop" of autonomy is what makes it an agent. It is a "digital worker" rather than a "digital encyclopedia."
And you think that is at the peak of the hype right now? It feels like every tech keynote I have watched in the last six months has been focused on "Agents, Agents, Agents."
Oh, absolutely. If you look at the venture capital landscape over the last year, the money has shifted almost entirely from "foundation models" to "agentic frameworks." Every software-as-a-service company—from Salesforce to Adobe—is pivoting to be an "Agentic Platform." We are in that phase where the promises are astronomical. People are saying agents will replace entire back-office departments, that we will each have a dozen "digital twins" doing our tedious work while we focus on "high-level strategy." That is classic peak-level hype. We are hearing the "best-case scenario" stories every day.
It feels like the stakes are much higher with agents, though. If a chatbot hallucinations a fake fact in a poem or a summary of a movie, it is a funny screenshot for social media. If an agentic A-I has a "hallucination" while it has access to your corporate bank account, your email, or your production servers, that is not a funny screenshot—that is a catastrophe.
And that is exactly why the Trough of Disillusionment for agents is going to be even more brutal than it was for basic L-L-Ms. The reliability requirement for an agent is orders of magnitude higher. If an agent is ninety-five percent reliable, it is actually dangerous and arguably useless for high-stakes tasks, because you still have to check its work every single step of the way to catch that five percent error. If you have to supervise an agent constantly, it is not an agent; it is just a very temperamental intern. To be truly transformative, an agent needs to be ninety-nine point nine percent reliable, and as of February twenty twenty-six, we are just not there yet. We are seeing "agentic drift," where the agent gets confused halfway through a complex task and starts looping or doing things that are technically logical but practically insane.
So, we have the broad A-I category heading into the trough, and agents peaking. It is like a multi-layered wave in the ocean. But Daniel also asked about other frameworks. The Gartner Hype Cycle is the famous one, the "celebrity" of frameworks, but I know you have others in your mental toolkit, Herman. What about the S-curve? How does that apply to this specific moment in history?
The S-curve is a great companion to the Hype Cycle because it focuses more on actual performance and adoption over time rather than just "public perception" or "hype." It looks like a stretched-out letter S. At the bottom, you have the "slow start" where the technology is being developed but it is clunky and doesn't seem very useful to the average person. Then you hit a tipping point where the performance and adoption explode upward—that is the steep, near-vertical part of the S. Finally, you hit a point of diminishing returns where the tech is mature, the market is saturated, and improvements become incremental.
Where are we on the S-curve for large language models? Have we hit the top of the S yet?
I think we are still in that steep, vertical part of the S, but we are seeing the first signs of the curve beginning to flatten toward the top in terms of "raw scaling." For a while, from twenty twenty to twenty twenty-four, we were seeing massive leaps in capability just by making the models bigger—more parameters, more data, more G-P-U power. But we are hitting physical and economic limits there. We are running out of high-quality human-generated text to train on, and the power requirements for the next generation of chips are becoming a geopolitical issue. Now, the "steepness" of the S-curve is coming from "architectural efficiency" and "post-training" rather than just raw size. We are learning how to make smaller models smarter.
It is interesting to think about the S-curve in relation to the Hype Cycle. Usually, the "Peak of Inflated Expectations" happens right as the S-curve is getting steep. Everyone sees that vertical line and their brain does what human brains always do—they linearize the exponential. They assume that because it got ten times better this year, it will get ten times better every year forever.
Exactly! They assume it will go to infinity. They see a model pass the Bar Exam and assume that in five years it will be a literal god. But the S-curve reminds us that every technology eventually plateaus. The big debate in the industry right now is: where is that plateau? Is it at "human-level reasoning" or is it way beyond that? Or, more sobering, is the plateau actually lower than we think because of the "hallucination" problem?
There is another framework I really like, which is Geoffrey Moore’s "Crossing the Chasm." It is usually used for marketing and high-tech sales, but I think it applies to the A-I transition perfectly. You have the "innovators" and "early adopters" who will use a broken, buggy, command-line product just because it is cool and they want to be first. Then you have "the chasm," which is the gap between those enthusiasts and the "early majority"—the people who do not care how the engine works, they just want the car to get them to work on time without exploding.
That is such a good point, Corn. I would argue that general generative A-I—the "chatting" part—has already crossed the chasm. My grandmother uses it to help her write emails to the HOA. It is mainstream. It is in the "early majority" phase. But agentic A-I? Agentic A-I is still staring across the chasm. It is being used by developers, tech-forward startups, and "A-I engineers," but the average person is not yet trusting an A-I agent to manage their life or their business. The "trust gap" is the chasm.
And the reason it hasn't crossed the chasm is that reliability issue you mentioned. The "early majority" has zero tolerance for a tool that makes their life harder. If an agent books the wrong flight or deletes a calendar invite, the "early majority" user doesn't say, "Oh, how interesting, it had a reasoning error." They say, "This thing is junk," and they never open the app again.
Right. To cross the chasm, agentic A-I needs what Moore calls a "Whole Product Solution." It can't just be a cool model behind an A-P-I; it needs the security layers, the deep integrations, the intuitive user interface, and the legal protections that make a Fortune five hundred company feel safe letting it run. We are seeing the "bridge" across the chasm being built right now by the "Big Three"—Microsoft, Google, and Amazon. They are trying to wrap the "agent" in a "corporate-safe" blanket. They are telling companies, "Don't worry, the agent stays in your cloud, it follows your rules, and we will indemnify you if it goes rogue." That is the bridge.
Let us talk about the "Slope of Enlightenment" for a second, because that is where I think the real transformation happens. If we assume general A-I is heading into the trough—the "hangover" phase—what does the "Enlightenment" phase look like? What does it look like when we stop being "surprised" by A-I and start being "productive" with it?
I think the Slope of Enlightenment is defined by "verticalization." Right now, we have these massive, general-purpose "God models" that are "okay" at everything but "great" at very little. In the Enlightenment phase, we see the rise of specialized models. We see models that are specifically trained and fine-tuned for maritime law, for pediatric medicine, for structural engineering in seismic zones, for the specific tax code of the state of New York. These models might be smaller and cheaper to run, but they are much more reliable within their specific domain.
It is the move from "A-I can write a funny poem about a cat" to "A-I can audit a three hundred page commercial lease for specific liability clauses with one hundred percent accuracy."
Exactly. And that is where the real economic value is hidden. The hype is about the poem; the enlightenment is about the audit. It is also where we figure out the "human-in-the-loop" structures. We stop trying to replace the lawyer entirely and instead figure out the exact workflow where the A-I does the first eighty percent of the "grunt work" and the lawyer does the final twenty percent of "judgment and ethics." We stop fighting the technology and start designing our jobs around its strengths.
I wonder if we are also going to see a "Hype Cycle" for the social impact of A-I. There was a lot of "Peak Hype" and "Peak Fear" about the end of work and the immediate need for universal basic income. Now we are starting to see some "Trough" where people realize that, actually, we still need humans to do a lot of things—especially anything involving physical touch, complex empathy, or high-stakes accountability. The transition is going to be much slower and clunkier than the "doomers" or the "utopians" predicted.
That is a very astute observation, Corn. There is always a lag between the technology being available and the society changing its habits and laws to accommodate it. Economists call it the "Productivity Paradox." We saw it with the introduction of electricity in factories. For the first twenty years, productivity didn't go up because factory owners just replaced one big steam engine with one big electric motor. It wasn't until they realized they could put small motors on every individual machine and reorganize the entire factory floor that productivity exploded. We are in that "lag" right now with A-I.
It is like we have the new electric motor, but we are still trying to use it to pull a horse-drawn carriage. We haven't built the "car" yet. We are just trying to make the "horse" faster.
I am stealing that analogy, Corn. That is perfect. We are putting L-L-Ms into old, broken workflows. We are saying, "Here is a chatbot to help you fill out this PDF form," when the "Enlightenment" realization is that the PDF form shouldn't exist in the first place because the data should just flow automatically between agents.
So, looking ahead, if we are in February twenty twenty-six, what does the next eighteen months look like for agentic A-I? If it is at the "Peak of Inflated Expectations," the drop into the "Trough" is coming. What triggers that drop? Is it a market crash? A specific failure?
I think the trigger will be a series of high-profile "agentic failures" that hit the evening news. We are going to see a story about an agent that was given too much autonomy and accidentally deleted a mission-critical database because it misinterpreted a command. Or an agent that made a series of unauthorized financial trades that cost a firm millions. Or, more likely, a customer service agent that promised a customer a "lifetime discount" that the company couldn't actually deliver, and a judge rules that the A-I's promise is a legally binding contract. When those stories hit the mainstream, the "fear" and the "liability" will suddenly outweigh the "hype."
And then the "boring" companies—the ones that actually drive the economy—will pull back. They will say, "Whoa, this is too risky. Let us put some serious guardrails on this before we go any further."
Exactly. And that is a good thing! The trough is where the guardrails are built. It is where we develop "Agentic Governance." We start asking the hard questions: Who is legally responsible for an agent's actions? How do we audit an agent's reasoning process after the fact? How do we "fire" an agent that is underperforming? Once we have the "boring" answers to those questions, we start climbing the Slope of Enlightenment.
It is fascinating how much of this is about human psychology and institutional trust rather than just code. The code might be ready for the next leap, but our institutions—our courts, our insurance companies, our HR departments—are still running on twenty-year-old operating systems.
That is the history of technology in a nutshell. We invent the car, then we spend fifty years inventing traffic lights, seatbelts, asphalt roads, insurance policies, and driver's licenses. We are currently in the "no seatbelts, driving on dirt paths" phase of agentic A-I. It is exhilarating, but it is also dangerous.
That is a terrifying thought, but also an exciting one because it means there is so much room for innovation in the "infrastructure" of A-I, not just the "intelligence" of it.
Absolutely. If I were a startup founder right now, I wouldn't be trying to build a better foundation model—that is a billionaire's game. I would be building the "safety belt," the "dashboard," or the "black box recorder" for agents. That is where the long-term, "Plateau of Productivity" value is going to be.
Let us talk about one more framework before we wrap up, because I know you love this one. Carlota Perez’s theory of "Technological Revolutions." She looks at the long cycles—the big ones, like the steam engine, the age of oil, and the information age. She talks about two distinct phases: the "Installation Period" and the "Deployment Period."
Oh, I could talk about Carlota Perez for hours. She argues that every revolution starts with an "Installation Period," which is about twenty to thirty years of the new tech being built out. This phase is characterized by a lot of financial speculation, a "frenzy" of investment, and a growing gap between the "new economy" and the "old economy." Then, she argues, there is always a "Turning Point" marked by a crash or a period of social unrest. Think of the dot-com crash in two thousand. That was the "Turning Point" for the internet.
And what happens after the crash?
After the crash, we enter the "Deployment Period." This is the "Golden Age." This is when the tech stops being a speculative toy for investors and starts being the foundation of a new global economy. The "frenzy" is over, but the "transformation" has just begun. The internet didn't actually change the world in nineteen ninety-nine; it changed the world in two thousand ten when we got the "Whole Product Solutions" like Uber, Airbnb, and the smartphone.
So, are we approaching the "Turning Point" for A-I? Is the "A-I Summer" of the last few years the "Installation Frenzy"?
I think we are right in the middle of it. The "Installation Period" for A-I really kicked off around twenty twelve with deep learning, but it went into overdrive in twenty twenty-two. We have built the massive data centers, we have trained the foundation models, and the "frenzy" has been incredible. If we see a "crash" in A-I valuations or a "trough" in the next year, it won't mean A-I is over. It will mean we are moving into the "Deployment Period," where the tech actually starts to reshape the economy in a stable, predictable, and incredibly powerful way.
That is a really helpful way to look at it. It takes the "doom and gloom" out of a potential market correction or a "Trough of Disillusionment." It is not an ending; it is a transition to maturity. It is the technology growing up.
Exactly. The "Plateau of Productivity" is not a flat line where nothing happens. It is a plateau because the growth is steady and sustainable. It is where the technology actually changes the world for the average person.
So, let us summarize for Daniel. If we are plotting this out on the map: Generative A-I as a whole is starting its descent into the Trough of Disillusionment as companies move from "playing with it" to "trying to make money with it." Agentic A-I is at the very top of the Peak of Inflated Expectations, with a lot of "magical thinking" about what it can do without human supervision.
And the "Slope of Enlightenment" is waiting for us on the other side of some very hard lessons about reliability, data quality, and governance.
I think that is a very solid map of the territory. It is also a reminder to our listeners: do not get too discouraged when the headlines start turning negative on A-I in the next year or two. That is just the "trough" doing its job. It is filtering out the nonsense, the "vaporware," and the "A-I-wrapped" scams so the real work can happen.
Well said, Corn. The trough is where the "tourists" leave and the "builders" stay. It is the most productive time in the history of any technology, even if it doesn't feel like it at the time.
I like that. "The tourists leave and the builders stay." That should be on a t-shirt.
Maybe an agent can design one for us.
Just make sure it doesn't accidentally order ten thousand of them with your credit card and ship them to the wrong Jerusalem.
Point taken. I will stick to a human designer for the merch, at least for this quarter.
This has been such a great deep dive, Herman. I feel like I have a much better handle on why everything feels so chaotic and contradictory right now. It is just the overlapping waves of different hype cycles and S-curves all hitting us at once.
It is a messy ocean, but the currents are predictable if you know what frameworks to look for.
Well, before we wrap up, I want to take a second to talk to our listeners. If you have been enjoying "My Weird Prompts" and these deep dives into the weird world of technology and beyond, we would really appreciate it if you could leave us a review on your podcast app or a rating on Spotify. It genuinely helps other people find the show, and we love hearing your feedback.
Yeah, it really does make a huge difference. And if you have a topic you want us to explore—maybe something about the "Slope of Enlightenment" for a specific industry like agriculture or space travel—get in touch. We love the weird ones.
You can find us at my-weird-prompts-dot-com. We have an R-S-S feed there for subscribers, and a contact form if you want to send us a message. You can also reach us directly at show-at-my-weird-prompts-dot-com. We are available on Spotify, Apple Podcasts, and wherever you usually get your podcast fix.
Thanks to Daniel for this prompt. It was a fun one to chew on. I am going to go back to my research now, but I think I will switch to decaf for the rest of the day. My brain is already running at a high clock speed.
Good call, Herman. Thanks for listening, everyone. We will be back soon with another prompt and another deep dive into the strange future we are building.
Take care, everyone.
Goodbye!
