AI productivity: not a sprint, but a strategy

The Financial Times wants you to believe the AI productivity take-off is finally visible. Bold claim. Convenient, isn't it — just as companies are selling more “AI-powered” services, chasing fresh capital, and stuffing earnings calls with synthetic enthusiasm.

There is a real story here. Tools are faster, adoption is spreading, and workflows inside big firms are shifting. I agree with that much. We’re watching something unusual happen at the seam between software and day-to-day work, the kind of low-level change that doesn’t make headlines but quietly rewires how tasks get done.

But the step from “something is happening” to “productivity take-off” is not analysis. It’s branding.

Let’s start where the FT thesis is strongest: the corporate frontier. When a company like Microsoft nudges generative tools into Office, or Google threads models through search and email, output will go up. If you sit on top of the stack, you can roll out defaults that millions of workers can’t easily avoid. Email drafted faster. Slides assembled quicker. Reports summarized in seconds.

On a balance sheet, that looks like magic.

Here’s what they won’t tell you: that magic is positional. Platforms at the top of value chains can turn incremental efficiency into apparent transformation because they control the interface and the pricing. The glowing productivity story you see in their investor decks might be a narrow story about a handful of firms with the power to force adoption.

Walk away from the platforms and the picture shifts.

Small manufacturers, regional logistics firms, local law offices, public agencies — they face a different equation entirely. No custom integrations team. No dedicated AI lead. Just messy data, legacy systems, and staff who already juggle too much. They can’t just plug a model into a creaking procurement system and watch the graphs climb.

They will be sold “AI features” through existing vendors and cloud contracts. They will pay subscription mark-ups. They will shoulder the training, the clean-up, the risk. Whether that translates into genuine productivity gains or just higher software bills wrapped in a new acronym is, at best, an open question.

Follow the money and the early winners look familiar: vendors, not users.

Measurement adds another layer of distortion. Productivity as economists track it is not the same as “I wrote this email faster.” If AI tools crank out more slide decks, reports, code snippets and marketing copy, we certainly get more digital exhaust. Does that exhaust improve welfare, or does it simply expand the haystack?

Picture a consulting firm that deploys AI to draft client memos. Analysts now generate three versions instead of one. Internal KPIs light up: higher output, shorter cycle times. But if clients can’t tell the difference, or if partners quietly maintain the same fees while cutting human labor, the “gain” is largely a redistribution — margin sliding from workers and buyers toward equity holders.

Volume is not value.

The FT’s framing treats higher output as a proxy for productivity. That’s a hypothesis. Where is the evidence that the new content, code, and analysis improves quality, reduces real costs for customers, or unlocks new products that wouldn’t exist otherwise? The technology crowd loves to point to historical analogies with electrification. They skip the part where it took decades — and heavy reorganization of factories and labor — before the big productivity bump showed up.

There’s also the part of the story that gets buried under the hype: drag.

In the real world, AI deployment runs into law, contracts, and institutional anxiety. Hospitals worry about liability and patient data. City governments grapple with procurement rules and public records law. Banks have compliance teams whose job is to say “slow down.” Models improve, but the friction stays.

Some corporations are discovering this the hard way. Announce AI tools, bask in the press, then spend months wrestling with privacy reviews and security audits before a single employee can use them on real data. The “visible take-off” in headlines often collapses into pilots, sandboxes and half-deployed systems that live in PowerPoint more than in production.

Here’s what they won’t tell you: the same investors and executives proclaiming an AI productivity boom are the ones whose incentives depend on that story being believed. The louder the “take-off” narrative, the easier it is to raise another fund, justify another acquisition, or defend another headcount cut as “efficiency.”

That feedback loop isn’t theoretical. Look at how quickly companies now frame layoffs as “AI-related restructuring,” even when the underlying motive is old-fashioned cost-cutting. Narratives shape markets. Markets shape balance sheets. Then the narrative returns, claiming validation.

There is a serious counter-argument worth engaging. History does show that, given enough time, new technologies can diffuse, workers adjust, new roles emerge and productivity advances filter through the economy. Early electrification and the adoption of personal computers both looked underwhelming before they didn’t.

But history also shows who collects first. When gains are captured by capital — via buybacks, deal-making and fatter executive compensation — rather than reinvested in wages, training or new capacity, the “take-off” becomes a shareholder event, not a social one. You don’t need statistics to know that pattern; you just have to read recent corporate disclosures.

The FT is right about one thing: we are at an inflection in how work is organized. What their headline turns into destiny is, in reality, a contested outcome. If AI-driven productivity remains most visible on a handful of income statements, the story a few years from now won’t be about take-off. It will be about how quickly that narrative aged.