Biggest Sale Ever! 👉
No products in the cart.
Return to shop
Home > News > Why 90% of AI Firms See No Productivity Boom Yet
Workers swear AI makes them faster. The companies paying for it can’t find the payoff.
Companies are still waiting for the AI productivity boom they were promised, even as big tech capex climbs toward $800 billion in 2026. The reason: individual workers report real time savings, but roughly 90% of firms actively using AI say it had no impact on productivity over the past three years. Speed on one desk isn’t scaling to the whole company.
Picture an engineer who used to spend a week on a task now wrapping it up by lunch. Now picture her CFO squinting at the same quarter’s numbers, wondering where exactly that magic showed up. That gap, between how it feels and what it earns, is the whole story right now.
Business Insider laid out AI’s productivity paradox in stark terms: software engineers are shipping more code than ever, companies are spending enormous sums, and the economy-wide payoff still hasn’t landed.
The disconnect is measurable. A February National Bureau of Economic Research working paper, drawn from a survey of nearly 6,000 executives, found that about 90% of firms using AI saw no productivity impact over the prior three years. Meanwhile, leaders can’t stop talking about it. Business intelligence platform AlphaSense found “AI” and “productivity” appearing together in 637 earnings calls in the second quarter, up roughly 25% from a year earlier.
On the ground, the gains feel real. Software engineer Iren Azra Zou, at trucking logistics startup Double Nickel, says Anthropic’s Claude Code compresses work that once took a week into a single day. “It saves an insane amount of time,” she said.
But individual wins don’t automatically roll up. Amazon data scientist Sarthak Gupta is actually working longer hours in what he calls an “automation phase,” building pipelines and wiring AI into existing workflows. His bet is that the upfront cost pays back every time the work repeats. That is the optimistic read. The pessimistic one is that a lot of companies are stuck paying the upfront cost with no repeating payoff in sight.
McKinsey senior partner Alexander Sukharevsky calls it a “gen AI paradox”: pilots look great, individuals feel faster, but turning that into companywide gains is brutally hard. Uber’s COO said last month there was no direct correlation between more AI use and more useful consumer features.
That has triggered what the piece dubs a “tokenmaxxing” reckoning, where teams burn expensive tokens without moving the needle, just racking up bills. IE University professor Enrique Dans put it bluntly: “When a metric turns into a goal, it stops being a good metric.” The question shouldn’t be how many tokens you burned. It should be what you actually accomplished.
Here’s the pattern entrepreneurs should clock: a tool that boosts individual output is not the same asset as a tool that boosts the business. The value capture happens at the system level, not the task level.
Compare it to spreadsheets. Lotus 1-2-3 launched in 1983 and eventually became the procedural backbone of global finance. It didn’t happen overnight, and the early years looked a lot like this: obviously useful, not yet load-bearing. The lesson for operators is that AI is still novel software, not procedural infrastructure. The companies that win won’t be the ones with the flashiest pilots. They’ll be the ones that rewrite a workflow so the gain compounds every quarter, exactly the bet Gupta is making. If you’re deploying AI, the strategic question isn’t “is this faster?” It’s “does this become a permanent part of how we operate, or just a recurring invoice?” This is the same value-capture problem reshaping every AI business model right now (how business models are evolving with AI).
The honest counterpoint is that the boom might simply be late, not absent. JPMorgan’s chief US economist thinks LLM skills require less training than past tech shifts, which could mean “years, not decades” before gains show up. Moody’s economist Mark Zandi agrees the lift is coming, just slowly, likely not visible in the data until the late 2020s or early 2030s.
But there’s a darker scenario. A new Wharton paper warns that if the productivity boom never materializes, the current buildout could become “the largest misallocation of capital in history,” with some major tech firms risking real financial strain. With hundreds of billions in capex riding on a payoff that’s still theoretical, the margin for being wrong is thin. And companies are already citing AI in layoffs and hiring slowdowns, betting on gains that haven’t fully arrived.
Individually, often yes. Some engineers report cutting week-long tasks to a day. The catch is that those personal wins haven’t reliably translated into higher company productivity, revenue, or profit yet.
They’re betting the payoff is coming and don’t want to be caught flat-footed. Big tech capex is climbing toward $800 billion in 2026 on the expectation that today’s groundwork compounds later.
It’s when teams burn through expensive AI tokens chasing usage as a goal, instead of focusing on actual results. It turns a cost into a vanity metric and produces big bills without clear value.
Estimates vary. Some economists say late 2020s or early 2030s. A more optimistic camp argues it could be years, not decades, because AI is easier to learn than past workplace tech.
AI’s individual gains are real, but a tool that makes one person faster is not yet a tool that makes a company richer. The founders and operators who win this cycle will be the ones who stop measuring activity and start rebuilding workflows so the value compounds. AI is having its spreadsheet moment: clearly useful, not yet indispensable. The leap from novel software to procedural backbone is where the money is, and nobody has fully made it yet.
I love understanding strategy and innovation using the business model canvas tool so much that I decided to share my analysis by creating a website focused on this topic.
More About Me
Four months in, John Furner has reshuffled almost the entire top floor at Walmart. Here’s […]
A son’s idle curiosity, a free chatbot, and a 60-year-old secondhand find collided into a […]
Telling shoppers an item might sell out sounds like a sales killer. The data says […]
A phone shaped like a tin can. Hay nets for horses. A pill box people […]
Hark won’t say what it’s building. Nvidia, AMD, and Intel handed it $700 million anyway. […]
The coding tool that lived in developers’ terminals now wants a seat in your finance, […]
Beijing just made it harder for Chinese companies to spend money abroad, right when they […]
The world’s biggest brewer is spending Super Bowl money on soccer. Main Street America is […]
Business Model Examples and Types Using the Business Model Canvas for Detailed Analysis
Learn
Product
Company
Get daily, no-fluff insights on the latest business models, startup strategies, and trends delivered straight to your inbox.
Lost your password?
source
No products in the cart.
Return to shop
Home > News > Why 90% of AI Firms See No Productivity Boom Yet
Workers swear AI makes them faster. The companies paying for it can’t find the payoff.
Companies are still waiting for the AI productivity boom they were promised, even as big tech capex climbs toward $800 billion in 2026. The reason: individual workers report real time savings, but roughly 90% of firms actively using AI say it had no impact on productivity over the past three years. Speed on one desk isn’t scaling to the whole company.
Picture an engineer who used to spend a week on a task now wrapping it up by lunch. Now picture her CFO squinting at the same quarter’s numbers, wondering where exactly that magic showed up. That gap, between how it feels and what it earns, is the whole story right now.
Business Insider laid out AI’s productivity paradox in stark terms: software engineers are shipping more code than ever, companies are spending enormous sums, and the economy-wide payoff still hasn’t landed.
The disconnect is measurable. A February National Bureau of Economic Research working paper, drawn from a survey of nearly 6,000 executives, found that about 90% of firms using AI saw no productivity impact over the prior three years. Meanwhile, leaders can’t stop talking about it. Business intelligence platform AlphaSense found “AI” and “productivity” appearing together in 637 earnings calls in the second quarter, up roughly 25% from a year earlier.
On the ground, the gains feel real. Software engineer Iren Azra Zou, at trucking logistics startup Double Nickel, says Anthropic’s Claude Code compresses work that once took a week into a single day. “It saves an insane amount of time,” she said.
But individual wins don’t automatically roll up. Amazon data scientist Sarthak Gupta is actually working longer hours in what he calls an “automation phase,” building pipelines and wiring AI into existing workflows. His bet is that the upfront cost pays back every time the work repeats. That is the optimistic read. The pessimistic one is that a lot of companies are stuck paying the upfront cost with no repeating payoff in sight.
McKinsey senior partner Alexander Sukharevsky calls it a “gen AI paradox”: pilots look great, individuals feel faster, but turning that into companywide gains is brutally hard. Uber’s COO said last month there was no direct correlation between more AI use and more useful consumer features.
That has triggered what the piece dubs a “tokenmaxxing” reckoning, where teams burn expensive tokens without moving the needle, just racking up bills. IE University professor Enrique Dans put it bluntly: “When a metric turns into a goal, it stops being a good metric.” The question shouldn’t be how many tokens you burned. It should be what you actually accomplished.
Here’s the pattern entrepreneurs should clock: a tool that boosts individual output is not the same asset as a tool that boosts the business. The value capture happens at the system level, not the task level.
Compare it to spreadsheets. Lotus 1-2-3 launched in 1983 and eventually became the procedural backbone of global finance. It didn’t happen overnight, and the early years looked a lot like this: obviously useful, not yet load-bearing. The lesson for operators is that AI is still novel software, not procedural infrastructure. The companies that win won’t be the ones with the flashiest pilots. They’ll be the ones that rewrite a workflow so the gain compounds every quarter, exactly the bet Gupta is making. If you’re deploying AI, the strategic question isn’t “is this faster?” It’s “does this become a permanent part of how we operate, or just a recurring invoice?” This is the same value-capture problem reshaping every AI business model right now (how business models are evolving with AI).
The honest counterpoint is that the boom might simply be late, not absent. JPMorgan’s chief US economist thinks LLM skills require less training than past tech shifts, which could mean “years, not decades” before gains show up. Moody’s economist Mark Zandi agrees the lift is coming, just slowly, likely not visible in the data until the late 2020s or early 2030s.
But there’s a darker scenario. A new Wharton paper warns that if the productivity boom never materializes, the current buildout could become “the largest misallocation of capital in history,” with some major tech firms risking real financial strain. With hundreds of billions in capex riding on a payoff that’s still theoretical, the margin for being wrong is thin. And companies are already citing AI in layoffs and hiring slowdowns, betting on gains that haven’t fully arrived.
Individually, often yes. Some engineers report cutting week-long tasks to a day. The catch is that those personal wins haven’t reliably translated into higher company productivity, revenue, or profit yet.
They’re betting the payoff is coming and don’t want to be caught flat-footed. Big tech capex is climbing toward $800 billion in 2026 on the expectation that today’s groundwork compounds later.
It’s when teams burn through expensive AI tokens chasing usage as a goal, instead of focusing on actual results. It turns a cost into a vanity metric and produces big bills without clear value.
Estimates vary. Some economists say late 2020s or early 2030s. A more optimistic camp argues it could be years, not decades, because AI is easier to learn than past workplace tech.
AI’s individual gains are real, but a tool that makes one person faster is not yet a tool that makes a company richer. The founders and operators who win this cycle will be the ones who stop measuring activity and start rebuilding workflows so the value compounds. AI is having its spreadsheet moment: clearly useful, not yet indispensable. The leap from novel software to procedural backbone is where the money is, and nobody has fully made it yet.
I love understanding strategy and innovation using the business model canvas tool so much that I decided to share my analysis by creating a website focused on this topic.
More About Me
Four months in, John Furner has reshuffled almost the entire top floor at Walmart. Here’s […]
A son’s idle curiosity, a free chatbot, and a 60-year-old secondhand find collided into a […]
Telling shoppers an item might sell out sounds like a sales killer. The data says […]
A phone shaped like a tin can. Hay nets for horses. A pill box people […]
Hark won’t say what it’s building. Nvidia, AMD, and Intel handed it $700 million anyway. […]
The coding tool that lived in developers’ terminals now wants a seat in your finance, […]
Beijing just made it harder for Chinese companies to spend money abroad, right when they […]
The world’s biggest brewer is spending Super Bowl money on soccer. Main Street America is […]
Learn
Product
Company
Business Model Examples and Types Using the Business Model Canvas for Detailed Analysis
Learn
Product
Company
RECEIVE OUR UPDATES
The Biz Model Club
Get daily, no-fluff insights on the latest business models, startup strategies, and trends delivered straight to your inbox.
Get daily, no-fluff insights on the latest business models, startup strategies, and trends delivered straight to your inbox.
Lost your password?
source
Leave a Reply