AI Starts Optimizing The Economy’s Plumbing — 2026-10-04

Azeem's main example is electrification, but it's also a perfect analogy for AI.

He cites International Energy Agency data showing that electricity now powers forty-six percent of global GDP, even though it represents only twenty-three percent of final energy use.

That gap matters.

If you look only at final energy, the world seems barely electrified.

If you look at value creation, electricity is already carrying nearly half the economy.

Why? Because electricity is more efficient.

A gasoline car turns roughly twenty-five percent of fuel into motion.

An electric car can convert around eighty-five percent to ninety percent into motion.

More broadly, Azeem notes that a joule of electricity does about two point five times as much useful work as a joule of oil.

That's the analogy I want to carry into AI today.

If we measure AI by average consumer usage or by how many office workers have fully autonomous agents, adoption may look early.

But if we measure AI by its role in the highest value workflows of coding, finance, procurement, scientific research, customer operations, infrastructure planning, the transition may be much further along than it appears.

This is exactly what Aaron Levy posted on X on October fourth.

He said AI agent adoption is still very bimodal.

Coding and coding adjacent work have taken off while everything else is more uneven.

He also noted that AI is creating new jobs around applied AI engineering, enterprise deployment, and FDE style services.

That bimodality is not a contradiction.

It's how transitions happen.

Electricity didn't electrify every task evenly.

It first transformed the parts of the economy where its advantages were overwhelming.

AI is doing the same.

The question is not whether everyone uses agents equally.

The question is whether agents are beginning to optimize the economy's hidden bottlenecks, and one of those bottlenecks may be money itself.

Azeem's main example is electrification, but it's also a perfect analogy for AI.

He cites International Energy Agency data showing that electricity now powers forty-six percent of global GDP, even though it represents only twenty-three percent of final energy use.

That gap matters.

If you look only at final energy, the world seems barely electrified.

If you look at value creation, electricity is already carrying nearly half the economy.

Why? Because electricity is more efficient.

A gasoline car turns roughly twenty-five percent of fuel into motion.

An electric car can convert around eighty-five percent to ninety percent into motion.

More broadly, Azeem notes that a joule of electricity does about two point five times as much useful work as a joule of oil.

That's the analogy I want to carry into AI today.

If we measure AI by average consumer usage or by how many office workers have fully autonomous agents, adoption may look early.

But if we measure AI by its role in the highest value workflows of coding, finance, procurement, scientific research, customer operations, infrastructure planning, the transition may be much further along than it appears.

This is exactly what Aaron Levy posted on X on October fourth.

He said AI agent adoption is still very bimodal.

Coding and coding adjacent work have taken off while everything else is more uneven.

He also noted that AI is creating new jobs around applied AI engineering, enterprise deployment, and FDE style services.

That bimodality is not a contradiction.

It's how transitions happen.

Electricity didn't electrify every task evenly.

It first transformed the parts of the economy where its advantages were overwhelming.

AI is doing the same.

The question is not whether everyone uses agents equally.

The question is whether agents are beginning to optimize the economy's hidden bottlenecks, and one of those bottlenecks may be money itself.