← Back to Interests
AI & Autonomous Systems

Using AI for real workflows, not just demos

I’m interested in AI when it leaves the toy stage and becomes useful: learning workflows, owner outreach, inbox-driven automation, expense tracking, and small systems that reduce friction without pretending to replace judgment.

The arc so far

My AI workflow started with simple assistance and quickly became more practical. Instead of treating AI as a chatbot novelty, I began using it as a coordination layer: reading inboxes, organizing data, building flashcards, preparing outreach drafts, and updating spreadsheets with structured information pulled from real documents.

The important shift was this: AI became more useful when it was connected to my actual systems — email, documents, spreadsheets, storage, and recurring tasks — while still keeping me in the loop for review and final decisions.

1) Website publishing workflow

The website itself has become one of the clearest examples of useful AI collaboration. Instead of manually editing HTML every time I want to add a new section, publish a deck, or tweak the design, I can describe the change in plain English and turn it into a live update.

Idea / requestplain-English change
AI edits sitecontent + design
Review / tweakhuman feedback
Git pushlive website update

This is the kind of publishing workflow I always wanted: low friction, fast iteration, and still fully reviewable before anything goes live.

2) Anki decks from real German reading

One of the most useful workflows has been turning German articles into Anki decks. Instead of memorizing random vocabulary lists, I can feed in a news paragraph and extract the words, phrases, and structures that actually slowed me down while reading.

German articlereal reading material
AI extractionwords + structures
Anki deck.apkg package
Study + websitelearn and publish

The best part is that the deck reflects what I genuinely need to learn, not what a generic textbook assumes I need.

3) Expense tracking from inboxes into Google Sheets

The most practical automation may be financial: giving the system access to my inboxes so it can pull bills, invoices, and statements, then update my Google Sheets and archive source files into the right Google Drive folders.

Inboxesbills + statements
AI extractiondates + amounts
Google Sheetsexpense updates
Drive archivesource documents saved

This is especially useful because it turns messy personal admin into something closer to a repeatable bookkeeping pipeline. I still review important decisions, but I don’t have to manually retype every number.

4) Cliffs owner outreach: email drafts, then text-message workflow

Another recent example was the AOAO board-election outreach for whole-unit owners at the Cliffs. The AI helped turn a raw contact spreadsheet into individualized outreach drafts, then later into a more manual but still structured text-message workflow.

Owner listspreadsheet contacts
AI draftsemail + SMS copy
Review/sendhuman in the loop
Replies trackedfollow-up list

Email stage

Build personalized Gmail drafts by unit, keep everything in draft form, and let me review before anything goes out.

Text stage

Create click-to-open text-message pages so I can work through contacts one by one instead of blasting messages blindly.

Reply tracking

Match phone numbers back to owners, separate replies from non-replies, and keep the follow-up list clean.

That’s the kind of AI use I like: not vague “agentic” hype, but something that saves real time while preserving review and control.

What I’ve learned

The sweet spot is not “let AI run everything.” It is closer to: let AI do the repetitive reading, sorting, extracting, and drafting — then let the human decide what matters. That balance works well for language learning, homeowner outreach, and personal finance administration.

In other words: AI is most valuable when it helps me think and act faster inside systems I already care about — not when it tries to impersonate judgment it doesn’t actually have.