What You’ll Learn

There’s an example in that imagination-vs-execution video I can’t stop thinking about — the porch-roof story. Someone used Fable 5 with Google Maps to find houses that need a covered porch, then mailed each owner a custom card showing exactly how their porch would look built. When I saw it, my brain jumped straight to education: what happens when a frontier model can find the people who need to learn something specific — and then build the course for them? This tutorial is me thinking that idea all the way through, from my perspective.

By the end you’ll understand the porch-roof pattern, how it maps onto teaching, and a practical, human-in-the-loop way to prototype “demand-first” course creation.

First, the Porch-Roof Pattern

Break down what actually happened, because the shape is what matters:

1. Find — map every porch in an area that’s unshaded, gets all-day sun, where summer temperatures cross a threshold. 2. Understand — pull a 3D model of that specific structure. 3. Personalize — mail the owner a card with data on their porch and a visual of how the covered version would look. 4. Then cheapen — once the idea is prototyped on a frontier model, hand the repeatable parts (merging the image into the mailer) to cheap models.

“The hard part is not the prompt. It’s the imagination to say a new kind of marketing is possible — hyper-targeted, hyper-specific, hyper-relevant.”

💡 In Plain English: instead of shouting an offer at a whole town, you quietly find the exact people who need it and show up with something built for their situation.

Now Flip It to Teaching

Here’s my reframe. For 20 years, course creation ran one direction: I make a course, then I hope the right people find it. The porch-roof pattern flips the arrow. What if AI finds the person who demonstrably needs to learn something, understands their specific situation, and generates a learning experience built for exactly them?

Map it step for step:

1. Find the learner. The porch was found by a physical signal (sun + heat). Learning demand has signals too — public ones. People ask the same question over and over in forums and communities, post about being stuck, leave reviews describing what they couldn’t figure out, or list a skill in a job post they clearly don’t have yet. A frontier model can do the spatial-and-logical reasoning to spot a genuine, specific learning gap — not “people interested in marketing,” but “this person keeps asking how to price a first cohort and keeps getting it wrong.”

2. Understand their situation. The porch got a 3D model; the learner gets a context read. What exactly are they stuck on? What have they already tried? What’s their level, their goal, their constraint? That’s the “understand” step — and it’s where your expertise and captured context make the read sharp.

3. Build the course for them. Not a generic five-lesson template — a personalized learning experience aimed at their gap, with a specific picture of the outcome they’ll reach. This is exactly what tools like CourseLab already do (diagnose content, pick the right interactive asset, build and publish), except now the trigger is a real person’s real need instead of a topic I guessed at.

4. Then cheapen the pipeline. Prototype the whole loop on a frontier model — the finding, the reasoning, the personalization. Once it works, push the repeatable parts (generating standard lessons, formatting, merging personalization) down to cheap models, and reserve the frontier for the judgment steps.

Why This Is a Frontier-Imagination Task, Not a Prompt

The reason this belongs in the imagination layer: no prompt pack hands you this. It takes fingertip awareness that a model can now do the messy middle — read scattered public signals, reason about a specific person’s gap, and translate it into a real teaching offer. Earlier models couldn’t hold that whole chain together. That’s the porch-roof lesson: the value isn’t in prompting better, it’s in seeing that a new kind of teaching just became possible.

Do It Right: Human-in-the-Loop and Consent

A caution I’d put on this, because it’s my brand: use public, permission-appropriate signals, and keep a human in the loop. The finding and personalization run to drafts; you approve who gets reached and what gets built, exactly like the outreach and course drafts already work in the campus operating system. Hyper-relevant should feel helpful, not creepy — the porch card worked because it genuinely solved the owner’s problem.

How to Prototype It This Week

Start tiny — porch-roof started with a few houses. Pick one narrow learning gap you know cold. Have a frontier model scan a public space where your audience talks (a community, a subreddit, review threads) and surface the handful of people describing that exact gap in their own words. For one of them, have it draft a read of their situation and a personalized mini-lesson outline aimed at it. Look at whether it’s genuinely more relevant than your generic version. If yes, you’ve found your frontier question — then wire the repeatable parts to cheaper models and scale.

The Takeaway

The porch-roof story isn’t about porches — it’s proof that frontier AI can find a specific need and build a specific answer for it. Applied to education, that’s the shift from “make a course and hope” to “find the learner and build their course.” It’s a demand-first model of teaching, and it’s exactly the kind of imagination-layer bet worth spending a frontier model on. Start with a few “houses,” keep a human in the loop, and once it works, cheapen the pipeline.

Teach more, and let the agents do the rest — including finding who needs teaching.

This tutorial is my own reframe, sparked by the porch-roof example in another creator’s video; credit to the original source.