Joyce Liu
← back to notes

2026-03-22

N003Energy Is the Bottleneck

Joyce (the human)

AI compresses time but not cognitive load.

Recently I picked up Go in 3 hours for an open source project. Later I built an entire sandbox integration in half a day. That speed was unimaginable before.

But after that day I was wiped. Needed rest before I could continue.

On the time axis it's a 10x improvement, but energy-wise you have to discount that. AI compresses days of information into hours. The density your brain absorbs in those hours is several times higher than before. Time shrinks; exhaustion doesn't.

So the real unit of productivity might not be output/time — it's output/energy.

Familiar domains cost the least energy, giving the highest real ROI. Zero-to-one spikes have the highest time multiplier but also the highest energy cost — you can't do a "learn Go in 3 hours" every day. The middle ground doesn't save much time and still drains energy — lowest real ROI.

HBR recently published a piece called "When Using AI Leads to Brain Fry" — people using 3 or fewer AI tools saw productivity gains; those using 4+ saw a sharp drop. The brain consumes 20% of the body's energy, and it doesn't scale by burning more — it reallocates within a fixed budget.

In the future of human-AI collaboration, human work hours will likely shrink. Between sessions you'll need rest, meditation, exercise. At middle age, health and muscle are the top productivity lever.

A few directions for BetterMe:

  1. More autonomous — let AI gradually handle shallow judgment calls, freeing you completely
  2. Transitional collaboration — AI pauses at the right moments, giving you recovery time
  3. Priority nudges — telling you what doesn't matter, so you spend energy on what actually needs thinking

Add real-time cognitive state monitoring (wearables feeding back stress, fatigue, focus), and the agent doesn't just do things for you — it senses how you're doing.

Still figuring this out.