Everyone's Productivity Is Up 10x, So Why Isn't Any Company Worth More?

Deep thoughts on AI and aspiration —— ByteDance Deep Thinking Circle

There’s a phenomenon worth watching closely: AI has pushed productivity up a notch for everyone who knows how to use it, yet not a single company has become ten times more valuable as a result. Where exactly did all that saved time and extra output go?

Hebbia CEO George Sivulka recently wrote an article that cuts straight to the heart of this puzzle. His verdict: we’re repeating the mistakes of the 1890s electrical revolution, committing one of the most expensive errors in technological history. He dug up this chapter of history not for a history lesson, but because its parallels with today’s AI adoption are almost frightening.

We’ve Swapped Motors Before

Let’s get this history clear first—it’s the foundation of the entire argument.

In the 1890s, as electricity became widespread, New England textile mills quickly replaced steam engines with faster electric motors. Logic would dictate that stronger power should mean higher output. But over the next three decades, these electrified factories saw almost no increase in production. The technology far surpassed what came before, yet organizational methods remained unchanged.

It wasn’t until the 1920s, when factories completely redesigned their production processes—introducing assembly lines, equipping each machine with its own motor, and assigning different work to workers and machines—that electrification’s returns truly materialized. In Sivulka’s words, these returns came from redesigning how the entire institution and technology worked together; technology alone, or making any single machine spin faster, couldn’t deliver them.

In other words, the factories that switched motors first lost to those that later redesigned their shop floors.

Your Company May Have Just Swapped Motors

Transplant this story to today, and the state of most AI applications becomes clear.

Sivulka breaks this down into two concepts: Individual AI and Institutional AI. The former consists of tools that make individual employees more efficient—writing emails, editing documents, searching for information, generating content. The latter weaves AI into organizational processes and makes it directly accountable for business outcomes.

His observation: what gets repeatedly discussed and shown off today is overwhelmingly the former. Employees share in group chats how much time they’ve saved using ChatGPT, but this kind of “productivity maximization” has almost no impact on actual business output. This is exactly what swapping motors looks like: individuals are faster, but the factory’s output hasn’t changed.

There’s a simple but easily overlooked fact here: efficient individuals don’t equal an efficient company. Individual efficiency gains are additive; organizational restructuring can be multiplicative. Focus only on the former, and you’ll mistakenly believe you’ve already captured AI’s dividends.

Why Individual Productivity Doesn’t Translate to Company Value

Sivulka lists seven contrasts in his article. I’ll highlight the three most illuminating ones.

First is coordination. He offers an analogy: suppose tomorrow you double your company’s headcount with clones of your best employees. These clones each have their own temperament and opinions. Without management, clear responsibilities, and communication mechanisms, you won’t get double the productivity—just chaos. Thousands of agents work the same way. When every employee has their own usage patterns and prompt habits, and outputs don’t connect, the organization may look more efficient at the individual level but is actually treading water at the organizational level, or even canceling itself out.

Second is outcome orientation. He quotes a VC’s observation: ask any CEO whether the top priority is cutting costs or growing revenue, and nearly all will say revenue. Yet almost every AI product on the market sells “save time, save headcount.” Time savings are easily commoditized—everyone can claim speed. What truly accumulates value is tying technology to results and directly creating revenue. Selling tools and selling outcomes are two completely different businesses.

Third is initiative. He makes a pointed observation: prompting AGI is like connecting an electric motor to an old loom—the bottleneck is stuck at the organization’s weakest link, which is humans. People barely know what questions to ask, let alone when to ask them. AI’s most valuable work is precisely the work no one thought to request: finding risks no one flagged, counterparties no one considered, channels no one knew existed.

There’s Also a Hidden Cost: More Noise, Same Signal

Beyond these three, there’s an easily missed ledger entry. Individual AI makes content production extremely easy, but most of what gets produced is garbage.

Sivulka himself acknowledges this contradiction: he runs an AI company yet requires his executive team not to use AI in any final written product because he can’t stand the garbage content. He’s blunt about it—even people selling AI are guarding against the noise AI produces.

Generation itself is no longer the problem; the problem is picking the right option from a pile of choices. He uses private equity investment as an example: last year you might have had 10 deal opportunities on your desk; this year you get 50 in a quarter, each polished to perfection by AI, but you still have the same amount of time to judge them. Individual productivity amplified, noise amplified with it, and finding the signal became harder.

Some organizations, seeing garbage proliferate, ban AI output outright. This is an overreaction, but it exposes a fact: without someone specifically responsible for “filtering,” individual-level output amplification ultimately becomes an organizational processing burden. Productivity doesn’t magically turn into value—there are still several steps in between: filtering, coordination, and judgment.

But Let’s Be Fair

At this point, I want to balance the argument, because this line of reasoning has two places where it can be pushed too far.

First, Individual AI isn’t useless. Sivulka himself admits it’s the entry point for most enterprises’ first experience with AI and the starting point for driving adoption and establishing change management. Without individuals using it first, organizational-level restructuring has nowhere to begin. So this isn’t either-or, but a matter of sequence and priority.

Second, the electricity analogy has limitations. Factories took decades to go from swapping motors to redesigning shop floors because electrical technology iterated slowly, giving organizations time to adjust at leisure. But AI evolves monthly, and the window for redesigning organizations is likely much shorter. Taking it slow may mean missing the opportunity. This is my biggest concern when comparing to this historical period: history will repeat itself, but the pace won’t be the same.

The Real Bottleneck Is Organizational Interface

Stacking these layers together, my judgment is: AI’s value doesn’t accumulate at the individual tool layer—it only accumulates in redesigned processes.

This means the bottleneck was never in the technology, but in the “organizational interface”—whether your processes, coordination methods, and incentive mechanisms have been rewritten for AI. Buying tools is just the entry ticket; redesigning the factory is where profit gets generated. Many companies think they’re doing AI transformation when they’ve only completed procurement.

In terms of action, two things can be done immediately.

For managers, stop measuring AI’s effectiveness by “how much more efficient employees became”—that’s a motor-swapping metric. Replace it with three questions closer to output: Did it generate new revenue? Did it discover risks no one flagged? Did it force us to change an old process? If you can’t answer any of the three, that investment probably remains at the individual level.

For entrepreneurs, stop just selling “time savings.” Time savings get commoditized, and prices get driven to the floor. Move upstream and tie technology to a specific business outcome—identifying a customer worth pursuing, blocking a problematic transaction, closing an extra deal. Value accumulates where the money is.

Electricity has been around for a long time. The question is: who moves first to rearrange their shop floor.