AI PILOT SCHOOL · COURSE MATERIALS

← All materials

Your Computer Can Do That Now

A 75-minute lecture + Q&A — outline & speaker notes

Presenter: Jaede · Materials: deck.html (slides), handout.html (take-home), run-sheet.md (demo ops) Audience: competent adults running on pre-2023 computer habits. No technical background assumed. Total stage time: 68 minutes of content + 7 minutes of transition slack = 75, then Q&A.

0. Cold open (5 min)

  • Thesis, verbatim: "You're not behind because you're incapable. The interface to computers changed in the last three years, and nobody sent a memo."
  • Marketing frame: this is the next iteration of software engineering — product makers, not programmers. Tonight is not a coding class and not a hype talk. It's about leveraging technology to build and to get your time back.
  • Who I am: I run a small household publishing operation out of Bend — websites, research, recipes, games — and nearly all of it runs on the systems you're about to see. Everything I show you tonight is something I actually use; nothing was built for this talk.
  • Promise: five rungs on a ladder, a real demonstration at every rung, and the price tag printed on every label — privacy, money, and the ways this stuff fails. You'll leave knowing which rung is yours and what to do Monday morning.

Speaker notes: Open cold on the thesis slide before introducing yourself — let the memo line land in silence first. The room is full of people who quietly suspect they're behind; the first job is to convert embarrassment into curiosity. Don't oversell: the credibility of the whole evening rests on "I actually use all of this," so say it plainly and move. When you give the promise, hold up the handout so they know the map is coming and they don't need to take notes.

Rung 1 — ASK (8 min)

Claim: It's a thinking partner, not a search box.

Demo beats: 1. Paste a dense, real-looking document into a chat model — the sample insurance coverage-change letter (sanitized stand-in, see run-sheet). 2. Ask: "Explain this in plain English. What are my options, and what would you ask before agreeing to it?" 3. Follow up conversationally: "What's the catch in the section about water damage?" — show that it holds context and digs deeper, which no search box does.

Tradeoff card — HALLUCINATION: - It will sometimes state wrong things with total confidence. Not often. Not never. - Verify anything that has a name, a number, or a dollar sign attached before you act on it. - Use it to understand faster; keep deciding from the primary document. - How I decided: I treat it like a brilliant intern — it drafts everything, it signs nothing.

Relevance test: This rung is for you if you've ever signed something you didn't fully understand.

Monday action: Take the next confusing document that lands on you — a bill, a form, a policy letter — and ask an AI to explain it in plain English and lay out your options before you act.

Speaker notes: This is the rung most of the room has technically touched — many have "tried ChatGPT." The reveal is not that chat exists; it's the quality of the ask. Most people ask like it's Google — three keywords — and get Google-quality answers back. The demo's question is deliberately long and situational; point at it and say "the question is the skill." When the follow-up lands, name what just happened: it remembered the letter, it went deeper, you were having a conversation with a document. Deliver the hallucination card without apology — the intern line gets a laugh and it's the correct mental model.

Rung 2 — RESEARCH (10 min)

Claim: It reads the pile so you don't have to.

Demo beats: 1. Show the pile honestly: a folder of five real public PDFs — city budget documents, a utility rate study, an HOA reserve report (public documents, see run-sheet). 2. Open NotebookLM with the folder already loaded; ask for a five-minute briefing of what matters to a homeowner. 3. Click one citation — show that every claim links back to the exact passage in the source document. This is the difference between this rung and Rung 1. 4. Play ~30 seconds of the generated audio overview: the pile of PDFs is now a podcast for the drive to work.

Tradeoff card — YOUR DOCUMENTS GO TO A CLOUD: - Uploading a document means a copy leaves your control. For public PDFs, that costs nothing. - For private documents, know what you're trading: read the retention basics once, then decide by category, not document-by-document. - Never upload other people's private information to a convenience tool. - How I decided: public or low-stakes goes to the cloud; family and finances stay home — and "home" is a real option, which we'll come back to.

Relevance test: This rung is for you if there's a stack of reading you're pretending you'll get to.

Monday action: Pick the pile of reading you've been avoiding, load it into a research notebook, and ask for a briefing you can read in five minutes.

Speaker notes: The emotional core here is the pile — everyone has one. Let them see the file list and count five documents, hundreds of pages. The citation click is the trust moment; slow down for it. Say explicitly: "Rung 1 can make things up. This rung shows its receipts." The audio overview is the showman's beat — thirty seconds, no more, and watch the room while it plays; that's when phones come out to take pictures of the screen. Close the rung with the first appearance of the local-option thread: one sentence, planted, not explained yet.

Rung 3 — DELEGATE (10 min)

Claim: It can act on your behalf.

Demo beats: 1. Show a Telegram chat with my assistant bot — a real agent running on a $80 computer the size of a deck of cards, sitting on my shelf at home. 2. Send it, live: "What came into the inbox this morning?" — it reads a (seeded, stand-in) email inbox and summarizes. 3. Ask it to draft a reply to one of the messages. 4. Show where the draft landed: in the drafts folder. Not sent. It writes; I send.

Tradeoff card — AN AGENT WITH YOUR INBOX (the big one): - An agent that can read your mail can read all of your mail. Grant access like you'd hand someone your house key, because that's what it is. - Start drafts-only. Auto-send is a decision you make months later, or never. - Give the agent its own account with narrow access, not the keys to your main identity. - How I decided: mine runs on a separate account, it drafts and I send, and there's a written list of things it will never touch — money, passwords, anything about my family. - This one is a genuine tradeoff. Some of you will decide "never," and that's a competent decision, not a timid one.

Relevance test: This rung is for you if you answer the same message more than three times a week.

Monday action: Write down the three messages you answer over and over — that list is the job description for your first agent. Don't connect anything yet; just write the list.

Speaker notes: This is the centerpiece rung and the reason the talk earns trust: you are about to show the scariest capability and lead with its cost. The $80-computer detail matters — it collapses "this must be enterprise stuff" into "this is a hobby-sized thing a neighbor built." When the summary comes back, before anyone can be impressed, say "now here's what I had to decide to turn this on" and go straight to the card. Spend the most time on this card of any in the talk. The "never is a competent decision" line is load-bearing: it tells the room you're informing decisions, not recruiting. Local-option thread, second appearance: this agent's brain can be a model running in my house.

Rung 4 — AUTOMATE (10 min)

Claim: It can run without you.

Demo beats: 1. Show a folder with a long public report PDF in it — the kind of document someone gets paid to read. 2. Run the audit pipeline on it, one command, and let it work in the background while you talk. 3. Come back to the output: a structured findings report — page references, flagged numbers, a summary — generated while we were talking. 4. Show the scheduler log: the same job ran at 6 a.m. without me. Point at one line: "that timestamp is the whole rung."

Tradeoff card — CONFIDENTLY WRONG: - Automation doesn't fail loudly. It fails politely, on schedule, with perfect formatting. - The work doesn't disappear — it changes shape: from doing the task to checking the task. - Decide your checking regime before you trust the output: spot-check a percentage, and never let anything publish itself without a human gate. - How I decided: every automated job I run ends in a report a human reads. Nothing I automate sends, posts, or files anything on its own.

Relevance test: This rung is for you if part of your job happens on a calendar, not by choice.

Monday action: Name one chore you do on a schedule — a report, a status check, a roundup — and write down exactly how you'd check a machine's work if it did it for you.

Speaker notes: The rhythm trick of this rung: start the pipeline in beat 2, keep talking, and let it finish during the tradeoff card — the interruption ("it's done, let's look") dramatizes the claim better than any slide. The 6 a.m. log line is the actual demonstration; the findings report is just proof of quality. Be candid that this rung is where the "checking" honesty lives — the room's skeptics are right that automation gets things wrong, and you win them by agreeing and showing the regime. Local-option thread, third appearance: scheduled work is where home hardware quietly shines — no per-run bill.

Rung 5 — BUILD (15 min)

Claim: You can make software now.

Demo beats: 1. Ask the room for a one-page website idea (fallback topic staged, see run-sheet). Take the first shoutable suggestion that's small and legal. 2. Type a one-paragraph brief into Claude Code — plain English, no code words. Read it aloud as you type. 3. Watch it scaffold the site — narrate what it's doing in human terms ("it's making the files, wiring the pages together"). 4. Open the local preview. First contact: the thing exists. 5. Change one thing live by asking in English — "make the headline bigger and the background darker" — to prove iteration is conversation. 6. Deploy it to a real URL and put the QR code on screen. Audience loads it on their phones. The website they suggested twelve minutes ago is on the public internet.

Tradeoff card — COST & MAINTENANCE: - This is the rung that costs real money: subscriptions and usage add up to real monthly spend if you build often. - Software you build is software you own. It will break someday, and the person on call is you. - The model can't tell you what to build or why. Noticing the real problem is still entirely human — and it's a learnable skill. - Next step for this rung: the Small Batch Software workshop ($75) — a working session on exactly that skill: extracting a buildable spec from your own problem. Details at handsinjars.com. - How I decided: I build small, I build for us first, and I never build what I can't afford to maintain.

Relevance test: This rung is for you if you've said "somebody should make an app for that" this year.

Monday action: Describe one small tool you wish existed — one paragraph, no code words — and bring it to a workshop or hand it to an AI and see how far it gets.

Speaker notes: This is the fifteen-minute finale and the marketing frame made flesh: product makers, not programmers. Taking a live audience suggestion is the credibility gamble that separates this talk from every canned AI demo they've seen on YouTube — the run-sheet has the fallback if the room freezes or wifi dies. Narrate the scaffold like a cooking show, not a terminal session; nobody needs to understand the text scrolling by, they need to understand that you're not touching it. The phone-QR moment is the peak of the evening; let it breathe. Then, deliberately, bring the temperature down with the cost card — ending the demo on honesty instead of fireworks is the brand.

The Local Option (recurring thread — appears inside rungs 2, 3, 4)

Not a rung — a standing alternative that travels with the tradeoff cards:

  • The sentence, planted in Rung 2: "Everything I just uploaded went to a cloud — but there's a version of this where nothing leaves your house."
  • Second touch, Rung 3: the shelf agent's brain can be a model running on my own computer — a gaming PC, the kind in half the houses in this town.
  • Third touch, Rung 4: scheduled work runs free on hardware you already own.
  • Paid off in the close: one slide. Modern open models run well on a consumer graphics card. Private by physics, not by policy: your documents can't leak from a cloud they never went to. It's more work, it's a hobby today — but it's my default for anything about family or money, and if that's your line too, it's the rung-adjacent path to ask me about after.

Speaker notes: Resist making this a rung — it's seasoning, and it's the single most differentiated thing in the talk (nobody else's AI lecture runs models at home). Three one-sentence touches build the mystery; the close pays it off in under ninety seconds. The phrase to land: "private by physics, not by policy."

Close — Find Your Rung (10 min)

  • Handout walkthrough, front: the five rungs at a glance — claim, "for you if," Monday action.
  • The worksheet question, asked to the room and printed on the back: "What did you do this week that felt slower than it should be?" Write down three answers. Every answer lives on a rung.
  • The five Monday actions, restated in one pass (verbatim from the rungs above — one line each on the slide).
  • The local option, paid off (see thread above).
  • Build rung's next step: the $75 Small Batch Software workshop — see handout back.
  • Contact and everything shown tonight: handsinjars.com.
  • Final line, verbatim: "There's a second talk: how to make all of this compound. Tonight was the ladder; next time is flying lessons."

Speaker notes: The close converts revelation into homework. Make the room actually write the three answers — thirty seconds of silence, pens moving; it's the only participatory beat of the night, and it's the one that survives the drive home. When you restate the Monday actions, go fast — the handout has them; the slide is just proof they're small. End on the flying-lessons tease and stop talking; the tease works better without elaboration. Q&A follows immediately.

Q&A posture (notes only — not slides)

Expected questions and honest answers:

  • "Will it take my job?" — It takes tasks first. The people in trouble are the ones who don't know which of their tasks it takes. That's literally why this talk exists — find your rung and be the person who reorganized around it.
  • "Isn't it stealing from artists/writers?" — The training-data fight is real and unresolved; courts and contracts are working through it. Tonight was about using these tools on your own documents and problems, which is the least contested ground. If you make creative work for a living, the calculus is genuinely different — happy to talk after.
  • "What does it cost?" — Rungs 1–2: free tiers get you a long way; $20/month is the standard on-ramp. Rung 3–4: hobby money, mostly time. Rung 5 is where real monthly spend lives. Nothing tonight requires buying anything to start Monday.
  • "Is my data safe?" — Shortest honest answer: safer than a data breach headline suggests, less private than a filing cabinet. Category rule from Rung 2, and the local option exists precisely for the documents where the answer must be "yes, period."
  • "Where do I start?" — The rung where the worksheet put you. If you froze on the worksheet: Rung 1, Monday, first confusing document that hits your inbox.
  • "What about the second talk?" — Flying lessons: using AI to improve how you use AI — the method behind everything you saw tonight. Sign up for the list at handsinjars.com and you'll hear when it's scheduled. (No date promised.)

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