UMSI AI Clinic Bootcamp · Sunday, September 27 · 1:45–4:30 pm · LCSIB 2290

Automating Without Overwhelming Yourself

Most automation projects fail by trying to do too much. Today: find the smallest useful step of a workflow that is worth automating, have AI build it, run it on test data, and keep the judgment calls manual.

1
Should you automate it?
I talk: xkcd, return on investment, hidden costs1:45–2:00
- “Is It Worth the Time?” The original xkcd, and the math behind it
- Try it with a task of your own
- Return on investment, and the hidden costs: maintenance and over-engineering
- The value of manual processes
- Solving the wrong problem
- Today's thesis: the smallest useful unit
Open section 1 ↗opens in a new tab
2
Your projects through IDEA
Interactive: every team, its own client project2:00–2:30
- Your eleven projects come in three shapes: intake → triage, pile → catalog, many sheets → one table
- The IDEA framework: Identify, Document, Experiment, Adjust
- Walk your project through IDEA, one step at a time
- Leave with your smallest useful unit named in one sentence
Open section 2 ↗All eleven projects ↗
3
Tools of the trade
Chat assistants, Google Apps Script, Claude Code and Codex2:30–2:45
- Match the tool to the task
- What your agent can touch: read, draft, send, delete
- It asks before it acts, and it follows house rules you write down
- What leaves your laptop, and why today is all dummy data
- Which tool fits your project?
Open section 3 ↗
4
Three live demos
One per project shape, live in Claude Code2:45–3:05
- Intake → triage: every application gets a status
- Pile → catalog: versions, duplicates and personal information flagged
- Many sheets → one table: a join, not AI
Open section 4 ↗
Break3:05–3:20
5
Build your smallest useful unit
Hands-on, in teams, on dummy data3:20–4:05
- Pick the sandbox file that looks like your client's work
- Build the form, fill it with dummies
- Have AI write the script. Read its explanation. Then run it.
- When it hits a planted problem: a rule, or manual review?
Open section 5 ↗
6
Share-out and close
What did it hit? What stays manual?4:05–4:30
- Three or four teams: what did you automate, what did it hit, what stays manual?
- Keep it alive, and make it findable
- Before you automate the answer, check it's the right problem
- 4:30: Charles Kemp, “Discovering the Problem Behind the Problem”
Open section 6 ↗