In the past month, more and more people around me are exclaiming that the singularity has arrived. After LLMs and Agents matured, those who use them well have seen staggering efficiency gains.

How staggering? OpenClaw, an open-source project with nearly 400,000 lines of code and close to 200,000 GitHub Stars, was written by one person in three months. In 2024, a 20-person dev team working for three months might not have produced that much.

But I've also encountered a real puzzle: individual efficiency has indeed skyrocketed, but organizational output efficiency falls far short of expectations.

Carrying this puzzle, I dug into a piece of history — the electrical revolution. I discovered that over a hundred years ago, people fell into the exact same trap.

01. Decades Wasted in the Electrical Revolution

In the 1880s, electric motors began entering factories. People thought electricity was far more efficient than steam — the future surely belonged to electric power.

But by 1900, electric motors accounted for less than 5% of mechanical power in American factories.

Not because electric motors didn't work well. Most factories did what seemed perfectly reasonable: replaced the steam engine with an electric motor and changed nothing else.

Factory layouts stayed the same — still designed around a central power source, all machines connected by drive shafts and belts. Job divisions stayed the same. Management stayed the same.

Workers had it a bit easier, but overall output stayed flat.

Economist Paul David studied this phenomenon and called it the productivity paradox: the technology clearly arrived, but productivity just wouldn't rise.

02. Sound Like Your Company Right Now?

From my observation, what most companies are doing right now is essentially "giving employees an AI account."

Cursor subscriptions purchased, Claude memberships activated. Employees are indeed faster — code written faster, copy produced faster, slides done faster.

But then what?

To be blunt: a large portion of the time saved has turned into slacking time. Some people even got lazier, and those who don't use AI well have seen quality decline.

Job responsibilities unchanged, collaboration processes unchanged, reporting structures unchanged, performance evaluations unchanged. Individuals are faster, but the organizational bottleneck remains.

Before AI, writing code was expensive, producing designs was expensive, making slides was expensive. So the organization's core job was to distribute, coordinate, and manage execution tasks.

After AI, execution became cheap. One person plus AI can write what used to take five people.

So what became expensive?

I think it's figuring out what to do.

Defining a valuable problem, figuring out how to solve it, getting the team to understand and align. These capabilities have suddenly become far more important than execution.

But most organizations are still structured for "managing execution."

Layer upon layer of approvals exist to manage execution quality. Fine-grained division of labor exists to manage execution efficiency. Weekly and monthly reports exist to manage execution progress.

When execution is no longer the bottleneck, these aren't just redundant — they might be getting in the way.

Just like factories a hundred years ago: they swapped the power source, but the production line stayed exactly the same.

03. What Did Ford Get Right?

In the electrical revolution, the first to truly reap the dividends was Ford.

In 1913, Ford created the assembly line at Highland Park.

The line succeeded not because Ford had better electric motors. It was because he understood one fundamental thing: electric motors are fundamentally different from steam engines. Each machine can have its own independent motor — no need to connect everything through drive shafts.

In the steam era, factory layout was dictated by power transmission. All machines had to orbit the central steam engine, with drive shafts and belts delivering power. Machine placement wasn't based on efficiency — it was based on proximity to the power source.

Ford thought in reverse: since each machine can be independently powered, equipment can be arranged entirely according to the production process. So he designed from scratch: equipment along the workflow, conveyor belts connecting each station.

Result: assembly time dropped from 12.5 hours to 93 minutes. By 1923, Ford alone held over 60% of the American auto market, and more than half the registered cars worldwide were Fords.

Not because he used electric motors earlier. Because he redesigned the entire factory around the characteristics of electric motors.

04. Execution Is Cheap — Then What?

AI's amplification of individuals is already extraordinary. One person writing 400,000 lines of code. One person building a product that used to require an entire team.

But most organizations are still just "swapping the power source" — deploying AI tools while the organization itself doesn't budge.

What I keep thinking about is: AI has made "doing things" cheap, but "figuring out what to do" hasn't become cheap. In fact, because more things can be done, choices have gotten harder — this part has actually become more expensive.

The capabilities organizations truly need in the AI era have changed:

Before, execution was most important — can you build what's been decided?

Now, the most important thing may be defining problems — can you figure out what's worth doing and what the real problem is?

And alignment — can the organization help everyone understand why they're doing something, not just assign tasks?

05. My Own Explorations

Our team of 10 builds "Moonviz APP," and AI has been a huge help.

But I increasingly feel that whether saved "execution time" turns into "thinking time" is the key. Often it hasn't. Everyone works faster, but not necessarily thinks more clearly. Including myself.

So I'm planning a significant restructuring: eliminating front-end, back-end, and client-side technical divisions, and also eliminating the split between product and engineering.

Why? Because with AI augmentation, everyone can break through their original boundaries.

Specifically, I plan to manage the team by business objectives. Each person owns concrete business goals.

For example, the former "front-end engineer" would no longer be responsible for "completing requirements written by the product manager," but for "driving the app's sharing and virality." They'd research how competitors handle sharing, figure out what we should do, collaborate with client-side design, and build the sharing H5 pages or mini-programs themselves. Responsibility shifts from completing requirements to delivering business results.

The logic is the same as before: execution is cheap now; "figuring out what to do" is the bottleneck.

06. There Won't Be Another 40 Years

The electrical revolution took roughly 40 years from the 1880s to Ford's dominance in the 1920s. Change was slow back then — factories had plenty of time to adapt.

AI won't give us that time. The pace of technological iteration is on a completely different scale from a century ago. What used to be 40 years might now be one or two, or even six months.

The most dangerous state isn't "not using AI" — it's "using it, but nothing has changed." Because you think you've kept up, so you stop thinking about it.

Just like those factory owners in 1900 who installed electric motors and felt they'd embraced new technology. Twenty years later, when Ford crushed them, they probably hadn't realized what happened.

The tools have arrived. But tools solve execution.

The real bottleneck has quietly shifted.