AI PILOT SCHOOL · COURSE MATERIALS

Flying Lessons: Use AI to Improve Your AI

Lecture 2 — 75 min + Q&A — outline & speaker notes

Presenter: Jaede · Materials: TBD — Lecture 2 deck, handout, and run-sheet are a future iteration. Audience: attendees who have seen Lecture 1 ("Your Computer Can Do That Now") or get the 3-minute recap below. No technical background assumed.

0. Cold open (7 min)

Recap the ladder in three minutes, one slide, five rungs: ASK, RESEARCH, DELEGATE, AUTOMATE, BUILD. One line each — just enough that nobody who skipped Lecture 1 is lost.
The killer realization, verbatim: "You can use AI to improve your interactions with AI." The people whose AI results keep getting better aren't smarter — they run a loop.
Coin the role, once, here, and nowhere else again: the AI Pilot — someone who doesn't build the plane, they fly it. Pre-flight checklist, instruments, flight log, landing. Every station tonight is a piece of that cockpit.

Speaker notes: Open with the recap slide already up so latecomers and Lecture-1 veterans land in the same place inside three minutes — resist the urge to re-teach any rung, just name them. Then slow all the way down for the killer-realization line; say it exactly as written and let it sit a beat before the "aren't smarter" follow-through. When you coin AI Pilot, mark it verbally ("I'm going to use this word all night, so let's define it once") so the room knows it's a term, not a slide accident. Land on the cockpit image — pre-flight, instruments, flight log, landing — because that's the map for the next fifty minutes.

The Pilot's Loop (5 stations, ~50 min total)

Each station: concept, a live look at Jaede's real system, "your minimum viable version," and a tradeoff note.

Station 1 — FLIGHT LOG (10 min)

Claim: capture every session, or the loop has nothing to compound.

Demo beats: 1. Open the real running journal — entries submitted through a small local web app Jaede built for exactly this, not a notes app that got repurposed. 2. Read today's actual entry aloud: what happened, what was decided, what's next. 3. Point at the pattern, not the tool: a timestamped record of what you did and why, in your own words, the same day you did it.

Your minimum viable version: one running text file. Open it, add three lines — what you did, what you decided, what's next — close it. Under two minutes, zero purchases.

Tradeoff note: a log you don't maintain is worse than no log — it becomes a place false confidence hides. Write short, write often, and never write anything you'd mind rereading in a year.

Speaker notes: This station is the easiest to dismiss as "just journaling," so spend the first thirty seconds on why it's not: the flight log is the raw material every later station reads from. Show the real entry — mundane on purpose, a session that hit a snag and how it got resolved — because a polished example would teach the wrong lesson. The MVV has to look almost insultingly small: one file, three lines, done before their coffee cools. That's the point. Land the tradeoff note plainly; a log is only as good as the honesty in it.

Station 2 — MEMORY (10 min)

Claim: give the AI a plain-text substrate it can actually read, and curate what it's fed.

Demo beats: 1. Open the vault: a folder of plain .md files, readable in any text editor, no proprietary format, no account required to open it. 2. Ask an AI a question that depends on a decision made months ago — something nobody would remember off the top of their head. 3. Watch it find the right file and answer correctly, then show the one line at the top of that file explaining what it is and why it matters — the curation that made the recall fast.

Your minimum viable version: one folder of .md files, named plainly, one topic per file. Move three existing notes into it today. Under fifteen minutes, zero purchases.

Tradeoff note: more files isn't more memory — an uncurated pile is as useless to AI as it is to you. A short note at the top of each file, saying what it's for, is what turns a folder into a memory.

Speaker notes: The demo's job is to make "plain text" feel like an advantage instead of a downgrade — no app, no subscription, no lock-in, and it still works if every AI company on earth vanished tomorrow. When the recall lands, name what just happened: the AI didn't remember, the file did, and the AI just read well. The curation half of this station matters as much as the storage half — a vault of a thousand unlabeled notes is a haystack, not a memory. Keep the MVV honest about effort: fifteen minutes, not an afternoon of reorganizing.

Station 3 — WRAP-UP (10 min)

Claim: end sessions so the next one starts smart instead of starting cold.

Demo beats: 1. Show a real session handoff note: a short written record left at the end of a work session for whoever — or whatever — picks it up next. 2. Start a brand-new AI session with only that handoff note as input. 3. Watch it pick up the thread in seconds — no re-explaining, no lost context — and compare the timestamp to how long a cold start usually takes.

Your minimum viable version: a three-question end-of-session template — what did I finish, what's unresolved, what should the next session read first — answered in the last two minutes of any work block. Zero purchases.

Tradeoff note: skipping the wrap-up doesn't cost you today, it costs the version of you that shows up next time and has to reconstruct everything from memory. The three questions are cheap; the amnesia they prevent is not.

Speaker notes: This is the flight-log's mirror image — the log captures what happened, the wrap-up decides what the next session needs to know. The cold-open contrast is the whole demo: let the room feel how fast the new session gets oriented compared to a normal cold start. Be concrete that this is a habit, not a tool — three questions, two minutes, no software required. If anyone asks whether this only works with AI: no, it also fixes handoffs between two humans; AI just makes the payoff visible faster.

Station 4 — FRAMES (10 min)

Claim: repeatable asks beat one-off asks — build the request once, reuse it forever.

Demo beats: 1. Show a personal file of saved prompt templates and reusable request "skills" — the exact wording that reliably gets a good result for a recurring kind of task. 2. Make the same request twice: first as a loose, one-off ask; then using the saved frame. 3. Compare the two results side by side — same AI, same task, visibly different quality, because one request had been debugged in advance and the other hadn't.

Your minimum viable version: a personal prompts.md file. Write down the best wording you've found for one task you ask AI to do more than once. Under ten minutes, zero purchases.

Tradeoff note: a frame that isn't revisited goes stale as the tools change underneath it. Treat your prompts file like a living document, not a one-time trophy — the winning wording this month may need a tweak next month.

Speaker notes: The side-by-side is the entire argument — don't over-explain it, let the two outputs sit next to each other and ask the room which one they'd rather have gotten. The insight to land is that most people re-invent their request from scratch every time, when the good version of that request already existed in their own head; a frame just writes it down. Keep the MVV to a single task, single file — the goal tonight is the habit of saving a frame at all, not a complete library.

Station 5 — INSTRUMENTS & HANDS (10 min)

Claim: give AI tools and an always-on home, and it stops being a chat window and starts being a system.

Demo beats: 1. Walk through a small always-on hub: one dashboard where an AI agent has authenticated connections to a handful of personal systems — the journal, the vault, a scheduling connector — and can act across all of them from one place. 2. Trigger one small recurring task live and show it running end to end without re-typing context into a chat box first. 3. Be explicit about what's on screen versus what's deliberately not connected — no credentials shown, no client or organization named.

Your minimum viable version: none. This station is aspirational for most people in the room tonight, and saying so plainly is more useful than pretending otherwise.

Tradeoff card — ACCESS = RISK: - Every connection you grant an AI system is a key you've handed over — decide per-connection, not all at once. - Ask, for each thing you're tempted to connect: what's the worst outcome if this account were misused, and can I live with that being possible? - Start with read-only or draft-only access wherever a service offers it; add write access only after the read-only version has earned trust over time. - Keep a written list of what you will never connect — money movement, anything about your family, anything you couldn't explain to someone else in one sentence. - How I decided: the hub only holds accounts I've deliberately handed it, each with the narrowest access the task needs, and there's a standing list of things it will never touch.

Speaker notes: Lead with the honesty, not the demo — say out loud that most people have no minimum viable version here, and that's not a failure of the audience, it's just where the frontier currently sits. The walkthrough should feel less like a magic trick and more like opening a toolbox: here's what's plugged in, here's why, here's what isn't. Spend real time on the tradeoff card — this is the most consequential decision in the whole lecture and it deserves more air than any other card tonight. The line that should land: every connection is a decision, and "never" is as legitimate an answer as "yes."

Close (8 min)

The loop stated once: every session deposits into the system; the system makes the next session smarter. Compound interest on context.
Durability promise: the tools shown tonight are today's parts — journal app, vault, handoff notes, prompt file, hub — and any of them can be swapped out. The loop itself is the method, and the method is what survives when the tools change.
Take-home: the Pilot's Preflight Checklist, one page, one line per station. And where the fuller course and templates will live: handsinjars.com — no date promised.

Speaker notes: Bring the five stations back together in one sentence each — this is the moment the room sees it was never five separate tricks, it was one system with five parts. The durability promise matters more here than it did in Lecture 1: this whole talk teaches a method precisely so it doesn't expire when the specific apps do. Close by pointing at the checklist as the thing worth keeping, and end on handsinjars.com without promising a date — the invitation is to watch, not to wait.

Open source under CC BY-SA 4.0. · AI Pilot School