Your AI strategy isn’t “evolving.” It’s flatlining.
While you were busy chasing the next leaderboard update and arguing about which model has the best “vibes” for coding, the real world just checked the receipts. And the bill is staggering.
Yesterday, the National Bureau of Economic Research (NBER) dropped a bombshell that should have every C-suite executive sweating through their Patagonia vests. After three years and over $250 billion in global AI investment, the results are in: 89% of firms report zero measurable impact on productivity or employment.
“The emperor isn’t just naked; he’s bankrupting the wardrobe department.”
The “1.5 Hour” Delusion
The NBER study surveyed 6,000 CEOs across the US, UK, Germany, and Australia. The most damning statistic isn’t the lack of ROI—it’s the usage pattern. The average executive uses AI for a pathetic 1.5 hours per week.
That’s not a digital transformation. That’s a novelty hobby. It’s the equivalent of buying a fleet of Ferraris to drive them to the mailbox and back once a week, then wondering why your logistics costs haven’t dropped.
We’ve spent three years treating LLMs like a magic “Apply Intelligence” button. We’ve built thousands of RAG-based chatbots that do nothing but summarize emails that nobody wanted to read in the first place. We’ve automated the trivial while the core business logic remains trapped in legacy silos that are “too risky” to touch.
The Agentic Mirage
The current industry pivot is to scream “Agents!” at every problem. The narrative is simple: LLMs were just the brain; Agents are the hands. If the brain didn’t fix the ROI, surely the hands will.
But as I’ve argued before in The Gen AI Paradox: The ROI Trap, adding “agency” to a broken process just creates a faster way to make mistakes. Most “agentic” deployments today are just complex loops of polling and human-in-the-loop bottlenecks. We are building Rube Goldberg machines out of tokens and calling it “autonomy.”
If you are still stuck in Pilot Purgatory, it’s because you are trying to “fit” AI into your existing organization instead of rebuilding the organization around the Agentic Substrate.
Why You are Failing (The Brutal Version)
- You are Polling, Not Reacting: You’ve built agents that ask for permission every five seconds. If a human has to approve every step, you haven’t saved time; you’ve just added a supervisor role to an AI’s internship. Read my take on The Death of Polling.
- Data is Still a Mess: You’re trying to run a $100k/month inference cluster on top of a data lake that looks like a digital landfill. Agents can’t navigate a swamp.
- The “Seat” Obsession: You are still paying for licenses per human “seat” while the work is being done by tokens. This is a fundamental misalignment of economics that I dissected in The Death of the Seat.
The “BarraCUDA” Warning
While the enterprise is failing to find ROI, the technical “moats” are evaporating. Look at BarraCUDA—a solo dev just replaced the “massive engineering moat” of CUDA-to-AMD compilation with 15,000 lines of clean C.
The software giants are charging you a premium for “proprietary stacks” that are being dismantled by teenagers in their bedrooms. The value isn’t in the stack; it’s in the Action.
The Provocation
The NBER study is a wake-up call for the “AI Tourists.” If your AI strategy doesn’t involve autonomous execution—systems that move money, ship code, and close tickets without a 1.5-hour-per-week CEO “reviewing” the output—then you aren’t investing in the future. You’re just subsidizing Nvidia’s next quarter.
Stop asking what AI can do. Start asking what your company is willing to let go of.
Are you building a system of record, or a system of action? Because the market only cares about the latter.
Related Reading:
- The $250B Lie: Why CEOs Are Admitting AI Productivity Is a Myth — The raw data and CEO survey breakdown.
Source Material: