Don't Bet on AI's Productivity Boom

They say there's an AI productivity boom.

The Budget Lab at Yale says don’t count your productivity‑data chickens. Smart caveat. But the piece does something more useful than waving a red flag; it forces a more uncomfortable question: what are we actually measuring when we point to “AI” and point to “productivity”?

Who's keeping score?

The article is right to go after the headline claim that short‑term upticks in measured output equal an AI‑fueled surge. Measurement isn’t a neutral mirror; it’s a stitched‑together Frankenstein of firm surveys, price deflators, and industry codes built for a different economy. When you apply yesterday’s tools to today’s software‑saturated firms, you get distortion masquerading as precision. The Budget Lab makes that point cleanly: data can mislead when the signals and the scaffolding don’t match.

But here's what they won't tell you: not all productivity is born equal.

When gains are concentrated in digital‑native firms that can fling software across millions of users at almost no marginal cost, the numbers can look spectacular while most workplaces feel frozen in time. Sectoral heterogeneity isn’t a side note; it’s the story. Manufacturing, finance, health care, hospitality — they don’t absorb automation and AI on the same timetable or in the same way. The article gestures at this divide. It should be the central interrogation: which sectors are actually moving, and which ones are simply being averaged into someone else’s boom?

Follow the money.

If productivity growth is clustered inside a handful of platforms, the aggregate data will spin a tale of broad advancement even as wages, hours, and service quality in other sectors flatline. That’s not speculative fantasy; it’s a structural possibility baked into any tech wave that scales fast in a few places and barely penetrates others. Policy reactions — from central banks calibrating interest rates to governments designing training programs — hinge on whether those gains are truly economy‑wide or mostly confined to a narrow digital frontier. Treating headline productivity blips as generalized progress risks sending attention and resources down the wrong corridors.

Where the Budget Lab piece really earns its keep is on time horizons. Short bursts of efficiency can be nothing more than reassigning tasks, pushing workers harder during rollout, or cleaning up measurement so that existing outputs suddenly look “more productive” on paper. AI pilots and early deployments are perfect conditions for productivity illusions — cost cuts, one‑off process tweaks, novelty effects. They do not automatically translate into a new, steady growth path. The article’s warning against leaping from initial efficiency to durable, economy‑wide transformation is not technophobia; it’s basic pattern recognition.

Data is political.

The Lab reminds readers that data is produced by institutions with their own incentives. Firms trumpet efficiency gains. Vendors showcase glowing “case studies.” Consultants publish surveys that also happen to prime their sales funnels. Convenient, isn't it. Those incentives seep into which inputs get tracked, which outputs are priced, and how “quality” is adjusted. The article’s call for skepticism about headline stats should spur more forensic work: trace the sources, interrogate the classifications, pull apart the assumptions hiding inside those deflators and output indices.

There’s a second layer the piece only sketches: the gap between measured productivity and lived economics. You can have rising productivity alongside stagnant real wages and rising prices. When that happens, the gains are being captured somewhere other than the median paycheck. Labor’s bargaining power, corporate governance, tax rules — these are the gears that decide whether AI‑driven efficiency becomes shared prosperity or shareholder yield. The article is right to puncture the idea that productivity growth is automatically a social good. It just stops short of naming the institutional chokepoints where that transformation can stall.

This is where history taps you on the shoulder.

Earlier waves of workplace software — email, spreadsheets, enterprise tools — boosted measured output in some sectors and inflated valuations in others. They also left plenty of workers feeling more surveilled than empowered. The internet era created giants like Amazon that looked hyper‑productive, while brick‑and‑mortar retail looked increasingly sluggish in the same data tables. The narrative was “digital efficiency”; the lived experience for many was speed‑up, schedule precarity, and intense pressure on wages. AI threatens to replay that pattern at a higher resolution unless we’re precise about who counts as “productive” and who just absorbs the fallout.

A counter‑argument goes like this: too much skepticism will slow adoption and choke off a genuine technological boon. If every data point gets cross‑examined, investors will hesitate, executives will delay deployment, and we’ll miss a chance to lift growth. The article’s tone, some will say, feeds into a culture of techno‑pessimism.

But clear‑eyed skepticism is not an anti‑growth posture; it’s a strategy filter. If policymakers and investors know where gains are real, where they’re ephemeral, and who actually pockets them, they can design training, competition rules, and tax systems that convert raw productivity into something closer to shared welfare. That requires more than quibbling over the data series; it demands building better ones.

Here’s what the Budget Lab only hints at: the hardest work ahead is methodological. Rethinking industry classifications so software‑heavy firms aren’t jammed into outdated boxes. Updating deflators so digital outputs aren’t mispriced. Tracking distributional outcomes — who sees higher income, who gets displaced, who gains bargaining power — alongside the usual averages.

Those methods won’t be built by press release or vendor white paper. They’ll come from statistical agencies that can say no to corporate lobbying, academic teams willing to trash legacy categories, and independent labs that don’t have a quota of “success stories” to hit.

Follow the money. If the AI productivity boom is real, you’ll see it not just in national accounts, but in pay stubs, bargaining tables, and the quiet rewrites of the measurement manuals.