Start learning

For working professionals

Learning AI while holding down a full-time job

You have maybe thirty minutes most days, and willpower already spent on your actual work. That's enough — once you stop losing it to the two things that waste it.

Two ways the time gets wasted

The restart loop

You start a course, get busy for three weeks, come back, feel lost, and restart from the beginning to refresh. Six months later you've taken the introduction four times and never reached the middle.

The restart loop: start the course, life gets busy, come back lost, start over from page one, and round again. Below it, the fix: come back after three weeks, resume where you stopped, be confused twice and then fine. The loop Start the course Life gets busy Come back lost Start over at page 1 six months, four introductions, no middle The fix Come back after 3 weeks Resume where you stopped Confused twice, then fine
Restarting feels responsible and is what keeps the loop going. Resuming badly is what gets you past the middle.

So don't restart. Resume, badly. Pick up where you were even though you've forgotten things. You'll be confused for two sessions and then fine, because the review pulls back what you need.

Waiting for a clear week

The clear week doesn't arrive. Twenty consistent minutes a day beats a five-hour Saturday on a quarter of the hours, because spacing does most of the work and a marathon collapses all of it into one day. See Why you forget.

Which is good news: the constraint you actually have — small, frequent slots — is the better condition, not the compromise.

Two bars comparing what is left a month later from the same five hours of study: one long Saturday session leaves a short bar, twenty minutes on fifteen separate days leaves a bar more than twice as tall. what you can still recall a month later 5 hours, one day 20 minutes a day same total hours · dots are study days
Same hours on the calendar. The gaps between them are doing most of the work.

Where to spend limited hours

Spend them on skills that appreciate as AI improves rather than ones the next model release makes worthless. That's the filter the whole Toorova curriculum is chosen with, and it works just as well on a Tuesday evening.

Two lines as models improve over time. The worth of prompt formulas and tool-specific recall falls; the worth of verification, calibration and problem framing rises, and the two cross. how much the skill is worth verification and framing prompt formulas as the models get better →
The same release that makes your prompt tricks obsolete makes catching its mistakes worth more.
Depreciating

Memorized prompt formulas. A specific tool's menus. Syntax recall. Step-by-step procedures for tasks a model already does. All valuable in 2023, worth less every quarter.

Appreciating

Verification — checking output you couldn't have produced yourself. As AI does more of the work, an undetected error costs more and the ability to catch one gets rarer.
Calibration — an accurate sense of where this tool is reliable and where it's confidently wrong.
Problem framing — deciding what's worth asking for, and what "good" means here. Models answer; people decide what deserves an answer.
Knowing when not to use it — covered separately, because it pays back fastest.

A routine that survives a bad week

  1. Attach it to something you already do. After the morning coffee, on the commute, before the first meeting. A slot defined by an existing habit survives; one defined by intention doesn't.
  2. Make the minimum embarrassingly small. One lesson. On a bad day, one question. The small version keeps the thread, and keeping the thread is what prevents the restart loop.
  3. Do the review, not the new material, when you're tired. Retrieval is cheaper than comprehension and, at 9pm on a Thursday, more productive.
  4. Apply one thing at work each week. One. Use something you learned on a real task and notice where it fails on real data. That failure is worth more than the next three lessons.
  5. Don't track hours. Track whether you showed up. Hours reward marathons, which is the wrong incentive.

Do you need the math?

It depends on whether you want to use AI well or build it. To judge output, spot a failing system, frame problems and know when not to reach for it, you need mechanism rather than equations. To train models and work out why the loss isn't going down, you need the math and there's no way around it.

Most professionals want the first, assume they need the second, and postpone starting over math they were never going to use. If you're unsure which you are, start with the first — it's a prerequisite for the second anyway.

There's also a shelf of role-specific tracks, aimed at applying AI in a particular job and carrying their own credential. If your goal is your current role rather than a career change, that's the shorter route.

Next