The Trough of Disillusionment is a Product Launch
The AI revolution is plateauing. The gap between the frontier models has narrowed to a statistical tie. Every week, a new release claims a 2% improvement on a coding benchmark that 99% of the world accepts but never verifies. We are drowning in evals and parameters, yet user enthusiasm is flatlining.
My hypothesis is that this inevitable hangover is the Large Language Model’s “Trough of Disillusionment.” Investors worry this is where the hype meets economic gravity.
But we are reading the chart wrong.
The Trough of Disillusionment isn’t an economic failure. It is a functional graduation. It is the natural transition from a technology defined by the marvels it generates to one defined by the products it powers. It is the moment we stop asking “How smart is it?” and start asking “What can it do for me right now?”
The Benchmark Trap
Federico Viticci at MacStories recently captured this shift:
“Ultimately, choosing between any LLM at the frontier of AI right now is a highly subjective matter that comes down to cost, workflow, app ecosystem, design, and, yes, pure ‘vibes’.”
He is right. We have reached the point of diminishing returns on raw intelligence. The early pioneers, OpenAI and Google, remain locked in an arms race to build a smarter brain. They assume “General Intelligence” is the product. They bet that if they make the model smart enough, we will forgive the friction of using it.
But look at what keeps users around. ChatGPT’s stickiness lately isn’t its reasoning on the LSATs. It is the sheer magic of Sora’s video generation. Google’s Gemini isn’t winning (non developer) hearts with better Python code. It is winning them with “🍌 nano-banana,” and natural language image editing feature on the Pixel that (mostly) just works ™.
The interface is currently still the chatbox. But the utility is no longer in the chat. It is in the specific, narrow superpowers tucked inside the interface.
The Terminal vs. The GUI
This brings us to a dangerous misconception in Silicon Valley. The giants believe they are building the next browser (in some cases, literally). Inherent is a bet the “Chatbox” is the final form of computing.
They are wrong. They aren’t building the browser (even though they are :P). They are building the Terminal.
Interacting with an LLM via chat feels exciting in the same way interacting with terminal felt exciting in 1982. It is powerful. You can do anything if you know the magic words. But it is a command-line interface masquerading as a product.
History tells us that command lines are for pioneers. Interfaces are for everyone else.
This is where ideally, “Neo-Companies” enter. Cursor realized developers don’t want a chatbot in a sidebar. They want an IDE that rewrites code for them. Cursor is “AI-Native” not because it has a better model, but because it hides the model behind a workflow.
Meanwhile, the “Tech-Native” pioneers risk becoming infrastructure providers for the companies that eventually displace them. We are watching the transition from the era of Technical Marvels to the era of Product Reality. The Trough isn’t the end. It is the filter.
Note: there is an argument to be made that even the Cursors of the world haven’t really innovated on the interface.
So, here’s a consideration – what if – the Neo company isn’t defined by when it was founded, but by how it treats the technology?
The Interface War
The Neo-Company understands that in the transition from marvel to product, the user interface is the product. The model is merely the electricity.
Two distinct strategies are emerging.
The Sustaining Innovator: Google
Critics screamed that Google was moving too slowly. But Google realized something from the mobile transition: in the early days, integration buys optionality.
Google treats the LLM revolution not as a disruption requiring new behaviors, but as “sustaining innovation” that supercharges old ones. They aren’t forcing you into a new “Google Chat” browser. They are injecting intelligence into the Pixel camera to fix your photos or summarizing chains in Gmail. They are betting the GUI of the future looks like the apps you already use, just smarter. They are productizing the technology by making it invisible.
AI overviews and AI mode from Google are the clearest example of this trend.
The Disruptor’s Trap: OpenAI
OpenAI is still betting on the Terminal. They built the fastest, smartest command line in history. But it is still a command line. It requires the user to learn a new language (prompting), switch contexts, and copy-paste results. It is a destination, not a companion.
While they captured the cultural zeitgeist, they are vulnerable to anyone who takes their API and wraps it in a better workflow. OpenAI is following the Microsoft playbook by building the GUI (a literal browser) and has the best Mac integration. But they are surprisingly slow to create the novel experiences that define a new AI-native behavior (sora being an exception).
Vibes vs. Workflow
Nowhere is this split more obvious than in software development.
On one side, we have Anthropic’s Claude. It captured the “vibes” market. The claude-cli is beloved by developers because it feels like a superpower in the terminal. It fits the hacker aesthetic. It signals that you are close to the metal, communing with the machine. It is cool. But it is a distinct tool you must pick up and wield.
On the other side, we have Cursor. Cursor didn’t just build a sidebar chat. They forked VS Code and rebuilt the entire IDE around the LLM.
Cursor started by extending current behavior. They supercharged autocompletions. You hit Tab, and the code appears. You highlight a bug, and it fixes it in place. Now they are paving the way for novel behaviors. The Agent Manager is the first realization that the value isn’t in the conversation, but in managing the outcome of the conversation.
It is fascinating that both Cursor and new tools like Antigravity rely on the “Inbox” as a metaphor for managing this complexity. This signals that the chat isn’t the destination. The completion is.
While Claude wins the battle for developer mindshare (and weirdly Twitter screenshots), Cursor wins the war for developer flow. One is a technical marvel that makes you say “Wow.” The other is a product that lets you go home an hour early.
The Great Handover
We have seen this movie before. The history of computing is a history of abstraction. We start with the raw, messy technical marvel, and then we build layers to hide it.
In the 1990s, the protocols powering the internet (TCP/IP, HTTP) were the technical marvels. But the economic upside did not go to the researchers who wrote the protocols. It went to Netscape, Google, and Amazon. They built the interface. They determined the novel behaviors (browsing, searching, one-click ordering) that actually mattered.
The Trough of Disillusionment is the market realizing that raw intelligence is becoming a commodity. As the gap between models narrows, the model itself ceases to be the differentiator. It becomes a utility.
In a world where intelligence is cheap and abundant, the value is no longer in the generation of the token. The value is in the routing of the token.
The pioneers did the heavy lifting of training the models. But they are often too close to the science to see the product. The Neo-Companies view the AI as the engine. They understand the average user does not want to prompt a model. They want to finish a task. They don’t care about the benchmark scores of the underlying LLM any more than they care about the voltage running through their dishwasher.
If you find yourself bored by the latest release notes or unimpressed by a 2% gain in math reasoning, do not worry. You aren’t disillusioned. You are eager for the next phase.
We are leaving the era of the chatbot and entering the era of the interface. The Terminal days are ending. The GUI is waiting to be innovated.
The revolution isn’t over. It just finally looks like software.
#ai #aiStrategy #anthropic #artificialIntelligence #claude #cursorAi #developerTools #generativeAi #google #googleGemini #llm #microsoft #models #nvidia #openai #productManagement #softwareDevelopment #techTrends #troughOfDisillusionment #userExperience #uxDesign