You get 5 hours a day. AI uses the other 19.

2026-07-12 · Ascendy Engineering

TL;DR

The time a person can purely use in a day is, generously, about 5 hours (subtract sleep, meals, a day job, commute, breaks, bathroom). The other 19 hours were being wasted anyway.
AI uses those 19 hours. And it’s cheap — all the subscriptions together cost about $5 a day, a few hundred won per hour. If someone asked “would you pay a few hundred won for one extra hour a day,” most would pay far more. And this is 19 hours.
So AI isn’t a tool — it’s leverage, on the one axis no one can buy back: time.
But there’s a twist. It didn’t work from the start. Before I built a fulcrum — verification — it was negative leverage: it amplified my losses. Only after building the fulcrum did it flip positive.

About this piece. A first-person account pulled out in an interview. “Leverage” is my bet, not a verdict. The 5-hour math is for a typical office worker, and the subscription prices are mid-2026 and change often. Same vein as I didn’t design it — I delegated it and token usage is not productivity.

The one axis you can’t buy back: time

The word “leverage” has several senses. For me, one of them dominates why AI is leverage — time. You can earn money and grow skill, but no one can buy time back. And people use far less of it than they think.

Let’s subtract. A day is 24 hours. For a typical office worker:

Sleep, 6 hours → 18
Three meals, 2 hours → 16
Day job, 8 hours → 8
Commute, 1.5 hours → 6.5
Coffee and breaks, 1 hour → 5.5
Bathroom trips, 0.5 hours → 5

Even under the fairly harsh assumption that you do nothing else, the time you purely secure on a weekday is 5 hours — and that’s for someone whose commute is under 90 minutes. So of 24 hours, only 5 are truly yours to use, and the other 19 were being wasted anyway.

AI uses those 19 hours — and cheaply

This is where AI comes in. Put AI to work and you can use those 19 hours you couldn’t. Code runs while you sleep, while you eat, while you’re doing your day job.

Look at the cost. All these subscriptions together run about $150/month — Claude Max about $110 (that’s with Korea’s 10% VAT), plus ChatGPT and Gemini subscriptions at roughly $20 each. Divide by the day and it’s about $5. (Mid-2026, VAT-inclusive in Korea; prices change often.)

At $5 a day for 19 extra hours, that’s under 30 cents an hour — a few hundred won. If someone asked “I’ll get you one extra hour a day, will you pay a few hundred won,” most people would pay far more; time is that precious. And this isn’t one hour — it’s 19. There’s no reason not to.

In practice, I now hand tasks to Claude Code, Codex, and Gemini before bed. Code gets worked on while I sleep, and in the morning I check the PRs that came up and merge. Bathroom breaks, coffee breaks — work moves in all those gaps too.

But — at first it was actually a loss

Hearing only this, it sounds like “just subscribe to AI, then.” No. This wasn’t possible for me from the start either.

Early on it was closer to negative leverage. Code got implemented in the wrong direction, or turned into a mess. Then fixing it took even more time. An anxiety about code correctness always lurked, and in the end I had to hold and review all the output myself. A lever amplifies not only good judgment but bad results, equally. Pull the lever without a fulcrum and what grows isn’t the gain — it’s the loss.

Only after building the fulcrum did it flip positive

What changed. I spent about a week building a review harness (redteam). Once I started working with it, the game changed.

Claude writes the code, and Codex reviews it. In almost every case Codex caught the wrong parts sharply.
Even the fixing got automated. I no longer had to hand-correct everything it caught.
The harness parallelizes appropriately, accounting for task dependencies and priority.

The result: the time I spend on verification dropped dramatically. And efficiency and code quality rose together. Where I used to hold all the output out of anxiety, a trustworthy verification layer now does that job.

This is the crux. The fulcrum of the leverage is verification. When the fulcrum is low, the 19 hours become a pile you have to clean up; when it’s high, the 19 hours become output as-is. That week of investment was spent raising the fulcrum.

So my 5 hours direct 19 machine-hours

So here’s what my day looks like now. My 5 hours of judgment direct 19 hours of machine work. I set the direction, check the results, and delegate the rest.

This morning was like that. All I did on waking was read the Codex review on a PR that came up overnight and merge it, then read our blog’s visitor report over coffee. That was it. It genuinely feels like working like a company’s owner — not doing everything myself, but sitting where you give direction and look at results.

Of course there’s a ceiling. Those 19 hours aren’t infinite — the ceiling is as much as I can verify and merge. So the higher I raise the fulcrum (verification capacity), the more of the 19 hours I can turn into output.

And that’s why the gap widens

Here comes the second sense of leverage — the gap. Same 24 hours, same $5 a day, and yet the result differs completely from person to person.

For someone who built the fulcrum, AI is positive leverage — 5 hours work like 24. For someone pulling the lever without a fulcrum, it’s negative — cleaning up what raw AI spat out eats another few hours. So the gap between them doesn’t narrow over time; it widens. AI being handed out equally doesn’t make results equal — it tilts toward whoever built the fulcrum.

Takeaways

AI is time leverage. It stacks the wasted 19 hours onto your 5 — at a few hundred won an hour.
But build the fulcrum first. If verification isn’t trustworthy, the 19 hours are a pile to clean up — that’s negative leverage.
The fulcrum is verification. I spent a week building a verification harness, and only then did the lever flip positive.
The gap widens. Same AI, same $5 a day, and the result tilts toward whoever built the fulcrum.

Honestly this too is my bet, not a verdict. But see AI only as a “tool” and you stop at some-cents-an-hour convenience; see it as “leverage” and you start asking where to place the fulcrum. That difference, I think, is what separates a 5-hour day from a 24-hour one.

Authorship & citation: Written by Ascendy Engineering; quotable with attribution. Found something wrong? Let us know via a GitHub issue.

Tags: ai, leverage, productivity, agents, career, opinion