Reserve Bank Governor Michele Bullock recently stepped up to a microphone and voiced what every traditionalist executive wants to hear: artificial intelligence looks suspiciously like a bubble, and national productivity metrics refuse to budge.
It is a comforting narrative for people who measure progress by paper clips counted and spreadsheets filled by hand. It lets legacy institutions off the hook. It suggests that if we just wait out the hype cycle, things will go back to normal.
I have spent the last three years advising enterprises sinking millions into machine learning integrations. I have watched boards panic, startups burn through venture capital on vanity chatbot wrappers, and legacy conglomerates fumble their first automation deployments.
The panic is real. But the diagnosis is entirely backwards.
Calling artificial intelligence a bubble because national statistics do not show an immediate spike in output is like judging the impact of the internet in 1994 by looking at fax machine sales. The metrics we use to track economic output were built for an industrial assembly line. They cannot measure the value of work that no longer needs to be done.
The Measurement Illusion
Economists love spreadsheets. Gross Domestic Product measures transactions, hours worked, and physical units produced. When a software engineer uses automated coding assistants to build a microservice in two hours instead of two weeks, traditional accounting registers that as fewer billable hours.
Fewer hours worked often translates in official reports to flat or declining productivity if output volume stays constant. That is not a failure of technology. That is a failure of imagination in how we calculate value.
Imagine a scenario where a mid-sized financial advisory firm deploys automated document parsers. Previously, ten junior analysts spent forty hours a week cross-referencing mortgage disclosures. Now, one senior analyst spends two hours reviewing flagged exceptions.
Output volume of reviewed documents remains identical. Labor hours drop by ninety percent. Under standard macroeconomic surveillance, productivity growth appears stagnant or mixed because the input hours plummeted. Yet the firm's cost structure collapsed, error rates vanished, and capital was redirected toward high-margin advisory services.
The productivity explosion is happening. It is simply hiding in the margins of legacy accounting models that cannot comprehend a world where marginal labor costs approach zero.
The Bubble Fallacy
Skeptics point to sky-high capital expenditures on data centers and silicon chips as proof of irrational exuberance. They compare current market valuations to the dot-com crash of 2000.
The comparison reveals a fundamental misunderstanding of what is being built. In 1999, pet food delivery startups with zero revenue went public at billion-dollar valuations. Today, the primary buyers of compute are generational infrastructure monopolies generating hundreds of billions in free cash flow, building out the foundational compute grid of the next century.
Call it a bubble if it makes you feel safe. But do not mistake market volatility for lack of utility.
The companies failing to extract value from machine learning share a common trait: they are trying to automate bad processes. Giving an incompetent organization a faster engine does not make it a race car. It just helps it crash into the wall faster.
I have seen companies blow seven figures on custom language models to summarize internal meeting transcripts that nobody read in the first place. They blame the tool when employee engagement does not soar. The software did not fail; the organizational discipline did.
The Uncomfortable Truth About Implementation
Let us be completely transparent about the downside of this technological shift. My contrarian stance is not that artificial intelligence is a magic wand that fixes broken business models. In fact, most deployments fail.
They fail because executives treat advanced probabilistic engines like plug-and-play Microsoft Office updates. They buy enterprise licenses, drop them onto employees' laptops with a twenty-minute tutorial video, and wonder why nobody uses them outside of drafting birthday emails.
True adoption requires tearing down existing operational workflows and rebuilding them around asynchronous intelligence. It demands that managers stop managing tasks and start managing outcomes. Most organizations are structurally incapable of this pivot because their entire middle-management layer exists solely to pass status updates between silos—tasks that automated workflows now execute in milliseconds.
When you automate middle management, you threaten the social structure of the corporation. That resistance is why productivity gains are bottlenecked in boardrooms, not in server racks.
Stop Asking the Wrong Questions
People ask when artificial intelligence will finally show up in national statistics.
Stop asking that. It is the wrong question.
The right question is whether your business will survive the next twenty-four months while your competitors operate at a cost basis you cannot touch.
National productivity numbers lag because macroeconomics is an autopsy report, not a weather forecast. By the time government statisticians update their models to account for autonomous code generation, automated tax preparation, and synthetic data simulation, the old economy will be a historical footnote.
The institutional hand-wringing over productivity stagnation is a defense mechanism for the obsolete. Ignore the macro noise. Tear up your operational flowcharts. If your current business model can be fully executed by an intelligent agent working at the speed of light, you do not have a productivity problem.
You have a terminal relevance problem. Fix it before someone else does.
