The Final Roadblock to the AI Supercycle
We are at a crossroads. Agentic AI capabilities have grown exponentially, yet most people still interact with AI as if it were just a chatbot, and they seem perfectly satisfied with that. At the same time, the era of token maxxing is ending. Token routing is becoming the norm, creating the impression that the inference economy is already running out of steam. Meanwhile, the governance layer of AI is emerging as a structural layer of the AI map, exactly as I’ve been arguing for months.
As usual, if you follow the outside chatter, the narrative is that it’s all over: the bubble is bursting, token consumption has plateaued, and the AI boom is losing momentum. I don’t spend much time debating whether we’re in a bubble. Focusing on the short-term financials of the most consequential industry of our time misses the bigger picture.
My job as an analyst is to explain what comes next. And today I can tell you this with confidence: there has never been a wider gap between perception and reality in the AI industry. Beneath the headlines, AI continues to accelerate. What we’re witnessing is not the end of the cycle, but the final bottleneck before mass adoption.
This is why I’ve also spun off The Business Engineer into The AI Supercycle.
You have probably noticed a strange split in the AI story this year. The demos are extraordinary. Agents that write working software run for hours on their own and manage portfolios of parallel tasks. The benchmarks climb each quarter; every keynote promises an autonomous workforce around the corner.
And then you go back to your actual job. You still type into a chat window. The “agent” your vendor sold you needs hand-holding to file an expense report. The CRM, the spreadsheets, the ticketing system you actually run on haven’t fundamentally changed. The most useful AI in your week is still a glorified assistant you check turn by turn.
This split is usually explained away as adoption lag. The technology is here, the story goes; users and companies just need to catch up. That explanation is wrong, and the way it is wrong matters — because it sends money and attention to the wrong places.
The split exists because three things that look like separate problems are actually one problem wearing three costumes:
Agents that don’t quite work yet at the frontier of real-world tasks
Users who won’t leave the chat window for anything that matters
Traditional software the market keeps funding instead of agent-native replacements
These are not three failures of imagination. They are three rational responses to the same missing piece of infrastructure.
And it’s about verification.
