Slow Travel × Learn AI — a real case study from Uncle’s actual toolbox
Everything in the Beginner’s AI Series is real — the Okinawa apartment hunt in the lessons isn’t a teaching example, it’s my life. I’m leaving for a multi-year slow-travel journey across Asia, and I need to know what monthly apartments actually cost in dozens of cities I’ve never seen. Here’s the system AI and I built for it — and what it teaches about going from “asking AI questions” to “having AI do jobs.”
Boomer’s note: “This is the graduate course. He did all eight lessons first, whether he knew it or not. So will you.”
The problem
Googling “apartment cost in Naha” gets you a blog post from 2019 and a listing site in Japanese. I needed current monthly prices for a whole watchlist of neighborhoods across Japan, Korea, and eventually Southeast Asia — refreshed regularly, comparable, and honest about what’s an estimate versus a verified number. That’s not a question. That’s a job. Questions go to a chatbot; jobs need a system.
The system (in plain English)
One spreadsheet is the boss. A master Google Sheet holds the watchlist — every neighborhood, every price found, with dates. It’s append-only: new information gets ADDED, old rows never get overwritten. That’s the factory rule again — you don’t erase the inspection log. You can meet the whole tracking system on The AI Company page.
AI does the sweeping. On a schedule, an AI assistant (Claude’s Cowork, doing the “workshop” jobs a chat window can’t) checks current listing prices for every active neighborhood on the watchlist and appends what it finds — with flags when a number looks off or a source is thin. One station, one task, in sequence: it’s Lesson 5, running without me.
I stay the QC. The system was built in stages with hard checkpoints: build it, run ONE live test on one town, then STOP for hand-verification before anything scaled up or got scheduled. Nothing automated until a human confirmed the first output by hand. If you’ve read Lesson 6, you know why that’s not paranoia — it’s the whole system working.
The rules that keep it honest
Scraped prices are a ceiling, not a promise. Every summary the system produces repeats this line: listed prices are what’s advertised, not what I’ll pay after negotiation, seasons, and reality. The numbers inform decisions; they don’t make them.
Some things never get automated. No logins, no bookings, no payments, no card accounts — ever. The AI researches; the human transacts. And when a neighborhood can’t be verified from good sources, it gets FLAGGED as unknown rather than filled with a confident guess. An honest “I don’t know” beats a fluent hallucination — that rule runs this whole website, including the dog.
What it took to learn
Honestly? The eight free lessons, applied. Asking what to ask surfaced things like key money and guarantors. SCOPE made the research prompts precise. Chaining broke “track Asia” into small stations. Verification became checkpoints. The only new ingredient is delegation — letting an AI run a proven routine on a schedule — and that’s a future lesson series of its own, once I’ve traveled with this thing and have real stories to tell.
🐾 You don’t build one of these. You describe one.
Read that system again and notice what’s missing: I never made the spreadsheet, never wrote the schedule, never touched a line of anything. I described the job out loud. Here’s roughly what I said — swap in yours.
“I want to track ___ across ___ over time. Build me the system: a spreadsheet that only ever adds rows and never overwrites old ones, and a job that checks each entry on a schedule and sends me a short summary. Show me the plan first and wait for my go. Test it on ONE entry before you scale it. When it’s running, explain what you built like I’m ten.”
Everything in this case study grew out of that paragraph. The two sentences doing the heavy lifting are “show me the plan first” and “test it on one before you scale it.” Those are what keep you the boss instead of the passenger.
Straight with you: this one isn’t free. A job that opens your own files and runs on a clock needs a paid account. Lessons 1 through 8 don’t — learn there first, and come back when you’ve got something worth automating.
The tracker started as one paragraph. Not one spreadsheet.
Boomer’s verdict
“A seventy-year-old with a robot research department and a strict no-touching-the-money policy. This is the most responsible thing he owns, including the passport. 10/10.”
🐾 Update, August 2026 — the tracker grew up
Since this was written, the system gained a standard (the Uncle Standard — four rules a place must pass), a hotel strategy for arrival weeks, and nine self-checks that audit its own numbers.
On its very first live test, those checks caught a scraper handing back confident, unverifiable data — and refused it. The machine now argues with its own sources.
Then came the real audit: I handed the whole rulebook to a rival AI and dared it to build the same tracker. It failed in ways mine never had — and three of those failures broke no rule, because the rule didn’t exist. Four new rules came out of that afternoon, plus one idea of theirs I stole outright. The story is Lesson 11 — The Second Opinion Machine.
It sweeps all 53 cities every month. The lessons that built it are free.
Start where I started: Lesson 1 of the Beginner’s AI Series — free, ten minutes, no experience needed.
Leave a Reply