01
Where does the data live?
Storage. The place facts are written down and can be found again tomorrow.
The brief
It arrived as three bullet points in a chat message
Not a spec document. Not a Jira epic. This is what real requirements actually look like.
Project: Stock Update & Alerts
- Weekly stock movement — which SKU came in / went out and how much, each week. (Currently someone updates an Excel file manually every Friday.)
- Low-stock alert — for best-selling SKUs, if remaining stock drops below 200 sq.m., alert so we can reorder.
- Dead-stock alert — any color/SKU with no movement for more than 4 months, alert.
Also: how do we connect Claude to Google Sheets?
Why this case is perfect for teaching: it is small enough to finish, real enough to matter, and it contains every classic decision a software project faces.
Where does the data live? What runs the logic? How do answers reach a human? What makes it run without one? Those four questions are the skeleton of every app you will ever be asked to build — and this course answers each one with an options table where the rejected options are argued honestly, and where the Against column of the winner is real.
The recurring theme: the skill is not writing code. The skill is thinking
clearly about the problem, and directing Claude well.
The decks
Three sessions
Session 0 stands alone — deliver it to owners and executives, no laptop required. Sessions 1 and 2 are hands-on and together fit in two days.
Session 0 · 1.5–2 hours
Think Like a Builder
The four questions behind every app, the discipline of not jumping to solutions, and four architecture decisions argued out loud.
- Every app is four questions
- Interrogating an incomplete brief
- The walking skeleton principle
- Push vs pull — why dashboards aren't alerts
- Coin-flips vs one-way doors
Open the deck →
Session 1 · half–full day
Setup & First Build
VS Code and Claude Code obeying you, the data-model decision that makes or breaks the project, and three business rules that meet data which fights back.
- Events vs snapshots — the one-way door
- Reading a genuinely messy spreadsheet
- Sample data with bugs planted on purpose
- The negative-stock row: our bug, or their data?
- The false dead-stock alert, fired before it's fixed
Open the deck →
Session 2 · half–full day
Connect, Alert, Automate
Google Sheets for real, alerts a tired human can act on at 18:05 on a Friday, a schedule that never forgets — and the post-mortem that makes this a case study.
- One client question, two different answers
- The service-account credential ceremony
- MCP: chatting with your own spreadsheet
- Alert design is UX writing
- Cron speaks UTC. Your client doesn't.
Open the deck →
Presenting? Inside any deck: ← → to navigate, N for
speaker notes, O to jump to a slide, F for full screen, ? for the key
list. Cmd/Ctrl + P prints the deck one slide per page.
The frame
Every app is four questions
The skeleton the whole course hangs on. Answer these before naming a single technology.
02
What turns data into answers?
Logic. The rules that convert raw facts into something worth knowing.
03
How do answers reach a human?
Interface or notification. An answer nobody sees is not an answer.
04
What makes it run without a human?
Automation. The trigger that fires whether or not anyone remembered.
And the architecture they produced
1 · DataGoogle Sheet
→
connectionSheets API
→
2 · LogicPython: 3 rules
→
3 · HumanEmail → LINE
⏱ Trigger: GitHub Actions cron, Friday 18:00 Bangkok — which is
0 11 * * 5, because cron speaks UTC.
What it teaches
The ideas that outlive the project
Four things went wrong across two days. Not one of them was a coding error.
Don't jump to solutions
Before saying "we should use X", ask what problem X solves and what else solves it. Every
decision arrives as an options table.
Cheap reversible defaults
Ambiguity is normal. You resolve it with a decision you can undo in an hour, not with four
weeks of meetings.
Events, not snapshots
Movements can rebuild any snapshot. Snapshots can never rebuild movements. It's why accountants
keep ledgers, not just balances.
Alerts must push
If a human has to go and look for it, it's a report, not an alert. Beginners always want to
build the dashboard.
Reality is broken
A warehouse showed −35 sq.m. of tile. Don't crash, don't silently clamp it to zero. Report it,
and keep running.
The worst bugs don't crash
The Friday job ran green every week, at 01:00 Saturday Bangkok time, with an empty report.
Nobody noticed for a month.
The decisions are the project
Delete the code and Claude rebuilds it from
CLAUDE.md. Delete the decisions and
you have nothing. Code is regenerable.
You own the truth
Claude wrote every line correctly. Every failure was a thinking failure — an assumption nobody
checked. That flinch is the job.
Meta
How this deck was made
The instructor wrote a CLAUDE.md-style brief: the curriculum architecture, every options
table's verdict, every speaker note's intent. Claude produced the artifact you're about to present.
The human did the thinking. The AI did the typing. That is the workflow the course teaches, applied one level up to the course itself.