Limitless AIOS.
01 /
Internal · Teaching research

How the best teach AI.

We watched 14 videos from the three biggest AI educators: Nate Herk, Jack Roberts and Brock Mesarich. Frame by frame, transcript by transcript, including all 8 videos inside Nate's 7 Day Challenge.

This is what makes their teaching stick for non-technical people, how they get information to be retained, and what our Foundations Course does with every lesson learnt.

Kauthar · 15 July 2026 · click through ›

Move 1 · The biggest one

Demo first. Define second.

Every single Nate video cold-opens on the thing already working: four agents running his real day, a reminder firing on its own. Only then does he explain a single term.

"Four different agents, all doing things for me in parallel. These are things I do every single day." Nate, before defining anything at all

The payoff earns the attention. The explanation spends it. People give a new idea about ninety seconds of faith, and he uses those seconds to show, not tell.

Ours: the course now opens with the Tuesday film. Watch a working day first, then every lesson explains what you just saw.

Move 2

One analogy per concept. Always.

Nothing stays abstract. Every concept gets exactly one everyday comparison, usually drawn on a hand-sketched slide, and it never changes:

  • Skills are "SOPs for your AI agents" (Nate) and "recipe cards" (Brock)
  • The context window is "a desk: only so much fits on it" (Brock)
  • CLAUDE.md is "an onboarding packet it re-reads every day and never forgets"
  • Model vs app vs you is "engine, car, driver" (Nate)
  • Connectors are "giving Claude hands" (Jack)

And they anchor to things people already own: "a skill is just what you put in a custom GPT, in a file." New idea, familiar shelf.

Move 3

Name the jargon, then defuse it.

Brock has a four-step reflex he runs on every technical word, and he never leaves one hanging:

  • Say the proper term once
  • "...which is basically just..." and the plain meaning
  • A household object: a phone, a recipe card, a desk
  • Show it in the real product within sixty seconds
"A markdown file, which is a fancy word for a very well-formatted document." Brock

The reassurance comes before the definition. That pre-empts the intimidation that makes non-technical people switch off.

Move 4

Keep the failure in.

When a demo breaks, Nate leaves it in the video and narrates the recovery in plain English. A dead API, a horrible first draft, duplicate rows: he shows the fix happening.

"It realised something didn't work, so it's trying another way. That's the beauty of agentic workflows." Nate, mid-glitch

Watching something break and recover builds more trust than a flawless run ever could. It also quietly teaches the one skill owners actually need: say what is wrong, let it go again.

Move 5

Volunteer the downsides.

The best videos have a dedicated honest segment: what it costs, what it cannot do, when it forgets, what happens if your laptop sleeps.

"Every ten minutes, that's more tokens, more tokens, and then context rot." Nate, on his own feature

Anti-hype is the credibility engine. The educator who names the limitation first is the one people believe about the capability.

Ours already does this: "Week one, it saves you minutes. Honest." is in the course on purpose.

Move 6 + 7

One question. One number.

Decision rules collapse to a single question. Two similar features? Nate does not compare specs:

"It's one simple question. Do you need help right now, or do you need help every day?" Nate

And demos end with the cost, in numbers. "That whole thing used 6% of my free credits." It answers the fear nobody says out loud: what is this costing me?

Move 8

The close is homework, not goodbye.

Nate's challenge does not end with congratulations. It ends with an assignment engineered to build the habit:

"Only use this. Just for the next week, try only using this. Take all your custom GPTs, put them in, and say: turn this into a skill." Nate, closing the 7 days

Adoption is engineered, not hoped for. Every lesson in our course now ends with one small action for the same reason, and Part II ends with the unattended week.

Retention · How it actually sticks

The memory toolkit.

  • Named frameworks as pegs: "the feedback cycle", "home, life, hands, grow", "the 7 levels". A name makes an idea storable.
  • Numbered progress, always visible: "concept 12 of 20", "skill 2 of 7". Nobody is ever lost.
  • Key takeaways at the end of every lesson: the whole lesson compressed to two lines you can re-find.
  • Tie-backs: every new concept explicitly connects to an earlier one, so it is one growing model, not thirty separate facts.
  • Quizzes with warm feedback: low stakes, wrong answers teach, nobody fails.
  • One running example improved across a whole lesson, instead of five disconnected demos.
The three educators

Nate, Jack, Brock.

Nate Herk is the calm peer. "I do not have a technical background" is his credential. Thirty minutes of ideas before a single install step. Why before how, always. The closest to our register.

Jack Roberts is the showman. Named ladders ("the 7 levels"), a hand-drawn card per level, and value framed in pounds saved per day. Brilliant structure, but hype pacing and paywall cliffhangers we will not copy.

Brock Mesarich is the translator. Nobody defuses jargon better. "Not a response, a deliverable." "Most people don't need a developer, they need an employee." That last line is practically our pitch.

Where we deliberately differ

What we will not copy.

They do

Teach beginners with the safety prompts switched off. Skip past key handling. Skip plan mode to look fast, then moralise after the mess. Hand-wave "your laptop must stay awake". Salt the teaching with sales plugs.

We do

Approvals stay on while learning. Keys handled properly as a named step. The plan and the yes-gate shown as the default, then one contained "here is what a snag looks like". Always-on hosting so nothing depends on a laptop lid. Teaching stays clean.

These are not style choices. For clients running real businesses on this, they are the product.

Where it all landed

The course, rebuilt on this.

  • Opens with the Tuesday film: payoff first, explanations after
  • One analogy per concept, reused for the whole course: the intern, the desk, the recipe card, engine-car-driver, vending machine or slot machine
  • Every lesson: words to read, a Remember block, one small action
  • Module checks: three warm questions, one at a time, wrong answers teach
  • Honesty built in: limits, the desk fading, "confidently wrong"
  • Part II is all homework: do it on your own AIOS, ending with the unattended week
Open the course ›
swipe ›