We are told that the fundamental purpose of software is to make us efficient, and that efficiency leads to macro productivity. With the explosive rise of AI, tech prophets promise historically unprecedented economic booms.

But the facts of economic history tell a completely different story.

Despite computing power increasing by over one trillion times since 1971, macro productivity growth has drastically slowed down, hovering near century lows. We have been seduced by an illusion: we mistake localized, task-level efficiency for true economic productivity.

The Productivity Paradox

In 1987, Nobel Laureate Robert Solow famously quipped:

“You can see the computer age everywhere but in the productivity statistics.”

Decades later, that paradox remains entirely unsolved. If you map the raw computational leap against actual economic output, the dissonance is staggering:

The Transistor Boom

In 1971, an Intel 4004 chip held approximately 2,300 transistors. Today, an Apple M2 Ultra contains approximately 134 billion transistors.

The Compute Leap

Microchip processing capability has scaled by a factor of roughly 10¹² since the early days of the microprocessor.

The Economic Flatline

Prior to the microchip era, from roughly 1920–1970, US labor productivity growth was comfortably above 2%.

Since the dawn of the personal computer and the internet, productivity growth has consistently struggled to maintain that pace.

The Brief Exception

There was a notable productivity boom between 1995 and 2004, driven in large part by the diffusion of information technology, the early internet, and organizational changes.

But outside that relatively brief period, software has arguably failed to produce the dramatic macroeconomic productivity gains that its proponents predicted.

Efficiency ≠ Productivity

We have to unpack a deep semantic divide.

Software can make an individual dramatically more efficient without necessarily increasing the amount of economically valuable output produced by the economy as a whole.

The "Excel Shock"

Consider the Excel Shock of the 1980s.

Digital spreadsheets automated tasks that previously required an army of accounting clerks. Calculations became faster, financial models became easier to build, and information could be manipulated almost instantly. But something unexpected happened.

Instead of simply eliminating administrative work or producing a proportional increase in corporate output, organizations began generating more financial models, more internal reporting, more forecasts, more dashboards, and deeper administrative loops.

We didn't necessarily produce more valuable economic product. We produced exponentially more information about the product.

The technology lowered the cost of producing analysis, and therefore increased the amount of analysis that organizations demanded. Efficiency created more work.

The Baumol Problem

True productivity growth requires a technology to revolutionize multiple industries simultaneously.

Instead, a huge proportion of IT investment has flowed into services:

• Therapy
• Legal counsel
• Healthcare
• Education
• Corporate management
• Consulting
• Financial services
• Administrative work

Many of these fall into categories associated with Baumol's cost disease, where output remains heavily constrained by human time, attention, and presence.

A Beethoven string quartet provides a useful analogy. A string quartet requires four musicians. The performance still takes roughly the same amount of time today as it did in 1826.

You can give the musicians better software, better communication tools, and better spreadsheets. But you cannot make the four-person performance take one-quarter of the time without changing the nature of the service itself.

Software can optimize the surrounding administration. It cannot necessarily make the underlying human service fundamentally more productive.

Where Did the Productivity Dividend Go?

Software was supposed to make us dramatically more productive. It was even supposed to help usher in something resembling John Maynard Keynes's prediction of a 15-hour workweek.

Instead, official working hours have broadly plateaued in many advanced economies, while the nature of work itself has changed dramatically.

The Endless Shift

Knowledge workers no longer truly "clock out."

Corporate communication channels—Slack, email, messaging apps, notifications—follow workers into their evenings, weekends, and vacations.

The office may have disappeared from the physical world.

But it has reappeared inside our pockets.

The Burnout Epidemic

The modern worker isn't necessarily working dramatically fewer hours.

Instead, work has become increasingly continuous, fragmented, and cognitively invasive.

The result is a strange paradox:

Technology has made individual tasks easier while making the overall experience of work more complicated.

Now Comes AI

As we enter the age of Large Language Models and autonomous agents, the technology ecosystem is doubling down on the same promise.

AI will make us more productive. But if a trillion-fold increase in computing power didn't automatically translate into systemic economic growth, why should we assume AI will?

The answer may lie in what AI actually makes cheaper.

If AI makes it nearly effortless to create:

• Text
• Code
• Images
• Presentations
• Reports
• Emails
• Marketing material
• Data analysis

then the marginal cost of producing each additional unit approaches zero. And when supply becomes effectively unlimited, the economic value of the individual unit tends to fall.

AI will undeniably act as a force multiplier. But what exactly is it multiplying? Potentially, raw digital output rather than economic value.

The Digital Overhead Hypothesis

The productivity dividend of our computing revolution didn't necessarily vanish into thin air.

The trillion-fold increase in computational power may simply have been absorbed by a massive expansion of digital overhead:

• More open tabs
• More software layers
• More notifications
• More data processing
• More dashboards
• More meetings
• More documentation
• More compliance
• More internal reporting
• More administrative steps between us and a finished task

We became incredibly good at producing and processing information. But information is not the same thing as economic value.

Software Didn't Make Us Elite Creators

Perhaps the most uncomfortable conclusion is this:

Software didn't necessarily make us elite creators. It made us hyper-efficient curators of an exponentially expanding digital environment.

We don't simply perform our jobs anymore.

We:

• Manage the tools that manage our jobs.
• Manage the data generated by those tools.
• Manage the notifications generated by that data.
• Create reports about the work.
• Create dashboards about the reports.
• Hold meetings about the dashboards.
• And then document the meetings.

Every layer can be individually justified. Every layer can make someone more efficient. And yet the system as a whole can become increasingly complex.

The AI Productivity Question

This is why the most important question about AI may not be:

How much faster can AI make us?

It may instead be:

How much economically valuable output does AI allow us to produce with fewer human hours?

Those are very different questions.

AI can make the creation of digital artifacts almost free. But making something valuable is a different problem.

The next productivity revolution will not come from generating more words, more code, or more images. It will come from eliminating entire categories of economically necessary human effort while increasing the value of what remains.

Until then, we should be careful about confusing more output with more productivity. Because history suggests that humans are remarkably good at taking every new efficiency gain, and finding an entirely new way to fill the time.