The Great Productivity Disconnect: Why AI Is Making You Faster But You – Good Machine

TL;DR

Three landmark studies — from MIT, Wharton, and Bain — landed within the same week, all converging on one finding: AI is dramatically boosting individual output while barely moving company-level productivity.
The MIT data is the smoking gun: AI coding agents boosted code volume by ~180% but shipped production code rose only ~30%. The gap between writing and shipping is where the real story lives.
The Wharton paper delivers the warning shot: tech companies are spending as if a productivity boom is guaranteed. If it doesn't materialise, "the current buildout will be the largest misallocation of capital in history."
90% of firms actively using AI reported the technology had no impact on productivity over the prior three years, per an NBER working paper surveying nearly 6,000 executives.
The individual vs. organisation gap is not a bug to be fixed with better AI — it's a structural feature of how knowledge work actually functions. The bottleneck is integration, not intelligence.

What Happened

In the space of roughly 72 hours this week, three separate research efforts — each from a different institution, each using different methodologies — landed on the same conclusion from different angles.

From MIT: A study across more than 100,000 developers found that AI coding agents boosted the volume of code written by roughly 180%, while the amount of code that actually shipped to production rose by only about 30%. Sarah Guo, founder of Conviction, summarised the finding in her newsletter The Untrainable: "As the model swallowed the part of software engineering you can best measure, we're relearning what many teams knew: engineering has always resisted measurement, and the most measurable parts may not be the only important ones."

From Wharton: Jessica and Jonathan Wachter published a paper finding that tech companies are spending as if they expect a productivity boom to materialise — but if it doesn't, "the current buildout will be the largest misallocation of capital in history." They warned that some major tech companies could risk bankruptcy if they don't quickly increase productivity.

From Bain: A study reported by Axios found that AI "is not paying off nearly as much as companies expected." The costs of AI are now; the profits are later — maybe. That "maybe" is what's making markets nervous, with the Nasdaq logging its worst day in 14 months on Friday.

These findings land on top of an NBER working paper from February that surveyed nearly 6,000 executives and found roughly 90% of firms actively using AI reported the technology had no impact on productivity over the prior three years.

This is not one study. This is a pattern.

What It Actually Means

The story here isn't "AI doesn't work." It very clearly does — at the individual level. Software engineer Iren Azra Zou told Business Insider that Anthropic's Claude Code helps her complete tasks in a day that used to take a week. "It saves an insane amount of time," she said.

The story is that individual speed and organisational throughput are different things, and AI is exposing — perhaps for the first time at this scale — just how wide the gap between them really is.

Think about what happens when a developer uses an AI agent to generate code. The code appears. It compiles. The tests pass. From the developer's perspective, the task is done. But the code still needs to be reviewed, integrated, tested against real systems, deployed through pipelines that may or may not be automated, monitored in production, and maintained for years. None of those steps got faster. Some of them — review and integration especially — may have gotten slower, because there's now more code to process.

This is the "gen AI paradox" that McKinsey senior partner Alexander Sukharevsky described to Business Insider: companies see promising results in pilot projects but can't turn isolated gains into companywide improvements. "Part of the challenge," he said, "is getting employees to both adopt the technology and learn how to use it effectively."

At Uber, COO Andrew Macdonald said last month there wasn't a direct correlation between increased AI use and "useful consumer features." His comments kicked off what Business Insider calls a "tokenmaxxing reckoning": workers burning tokens aren't necessarily bolstering company productivity — they're racking up sometimes massive bills.

Hype Deconstruction

Let's be clear about what this is not:

This is not evidence that AI is a bubble. The technology works. Individual productivity gains are real and substantial. The question is whether those gains can be aggregated.

This is not a call to stop using AI tools. If you're a knowledge worker who has found AI useful, you should keep using it. The individual gains are yours to capture.

This is not a prediction of mass unemployment. As Moody's chief economist Mark Zandi told Business Insider: "I don't think we're going to see mass layoffs or unemployment. We will see a lot of job loss in certain industries, but job gains in others."

What this is: the first clear signal that the easy part of AI adoption — making individuals faster — is done, and the hard part — making organisations faster — hasn't really started.

The Stakeholder Landscape

Individual knowledge workers are the clear winners right now. If you've figured out how to use AI tools effectively, you're producing more output in less time. The question is whether that translates into leverage (higher compensation, more interesting work) or just more output for the same pay.

Middle managers are in the hardest position. They're being told to adopt AI, to make their teams more productive, to justify the spending — but the tools that make individuals faster don't automatically make teams faster. The integration work falls on them, and it's largely invisible to the executives demanding results.

Executives and investors are facing a genuine crisis of measurement. If AI is making workers faster but companies aren't seeing the payoff, then either the metrics are wrong, the integration is failing, or the spending is indeed a misallocation. All three are probably true to some degree.

AI tool vendors have the most to lose from this narrative. If the story becomes "AI makes individuals faster but not companies," the enterprise sales pitch gets much harder. Expect a wave of "ROI frameworks" and "enterprise-grade" positioning in response.

Cross-Layer Implications

The measurement problem is a time-management problem. When a metric turns into a goal, it stops being a good metric — a point IE University professor Enrique Dans made to Business Insider about token-counting. "It's not about measuring people's productivity according to how many tokens they burn; that's absurd. The metric should be, 'what have you achieved?'"

The integration bottleneck is a coordination problem. AI can write code, but it can't negotiate with the team that owns the deploy pipeline. It can draft a memo, but it can't navigate the politics of who needs to sign off. These are the real time sinks in organisations, and AI doesn't touch them.

The capital allocation question is existential. If the Wharton paper is right — if the current buildout is the largest misallocation of capital in history — then we're not just talking about productivity. We're talking about a financial event that would reshape the tech industry.

What This Means for You

If you're an individual contributor using AI tools: Keep using them. The gains are real. But start thinking about how you demonstrate output, not activity. The person who ships more features, not the person who generates more code, is the one who gets promoted.

If you're a manager: Your job just got harder and more important. The value you create isn't in making individuals faster — AI is doing that. It's in doing the integration work that turns individual speed into team throughput. That means clearer handoffs, better review processes, and honest conversations about what's actually blocking work.

If you're an executive: Ask your teams not "how much AI are you using?" but "what has actually shipped faster because of AI?" If they can't answer the second question, the first answer doesn't matter.

If you're an investor: The framework Sarah Guo proposes is worth internalising. Ask whether a company's value depends on correctness that can only be verified inside private data, and whether that private environment requires access that takes years to obtain. Companies that satisfy both conditions are competing in what Guo calls "the untrainable corner" — territory where a smarter model is irrelevant because the bottleneck is permission, not intelligence.

The Uncertainty Ledger

What we don't know yet:

Whether the productivity gains will materialise at the company level with better integration, or whether the gap is structural and permanent.
Whether the Wharton paper's bankruptcy warning is a tail risk or a central scenario.
How much of the recent US productivity surge (above pre-pandemic trend since 2020) is actually attributable to AI versus remote work, job switching, and workforce composition shifts.
Whether JPMorgan chief US economist Michael Feroli is right that AI-driven productivity gains could materialise in "years, not decades" — or whether Zandi's timeline (late 2020s to early 2030s) is more realistic.

What would change the analysis:

A major company demonstrating clear, auditable company-level productivity gains from AI that aren't just individual anecdotes.
Evidence that the integration bottleneck is being addressed by tooling (rather than just better models).
A significant downward revision in AI capex spending plans from the hyperscalers.

Bottom Line

The most important time-management insight of 2026 isn't a new technique. It's that AI has made individuals dramatically faster while leaving organisations largely unchanged — and the gap between those two things is where careers, companies, and possibly entire investment theses will be made or broken. The spreadsheet didn't transform global finance overnight, and AI won't transform productivity overnight either. But the gap between writing code and shipping it — between generating output and creating value — is now the central problem of knowledge work. And nobody has solved it yet.

Sources:

Business Insider, "AI's productivity paradox" (June 10, 2026) — Tier 1
Forbes, "AI Coding Agents Write 180% More Code But Ship Only 30% More Software" (June 10, 2026) — Tier 2
Axios, "The business of AI's 4 harsh realities" (June 7, 2026) — Tier 2
NBER Working Paper (February 2026), survey of ~6,000 executives — Tier 1
Wharton paper by Jessica and Jonathan Wachter (June 2026) — Tier 1
MIT study via Sarah Guo / Conviction, The Untrainable (June 2026) — Tier 2
SHRM, "The Executive Download: HR Technology Trends, June 2026" — Tier 2