Structured Data vs Structured Interactions
Introduced by: Dave Kaminski (articulated) / Pete Kaminski (demonstrated)
Depth: Deep thread
The most quotable insight of the entire call, emerging from the contrast between Dave's experiences and Pete's methodology.
The Conversation
Dave had observed that Claude works brilliantly with structured data:
Dave: "One of the tricks is to really have your stuff originally in a structured data that's actually correct. And if you can do that, then this thing will just do anything for you. But if you have unstructured data, and you think that it's actually going to talk you through it, you're mistaken."
Pete agreed, then reframed the solution:
Pete: "So, it works really good with structured data. It's crap, and it runs through stop signs, and crashes into brick walls all the time when you don't have structure, right? So, the fix for that is to structure the way you work with it."
Dave captured this as a principle:
Dave: "You said something interesting, which was, you don't have to have structured data, you should structure the way you work with it. I don't know if that's in your guide, but that's, like, quotable, you know? And you should put that in bold."
Pete realized he'd never stated this explicitly:
Pete: "It's old for my class, because I never even considered the idea of having an unstructured way of Claude Code, so they never even learned how to do it an unstructured way."
Why This Matters
This insight bridges two worlds:
-
Dave's world: Messy, unstructured research data — historical postcards, handwriting samples, eBay thumbnails, overlapping date ranges. The data will never be clean. But you can still get great results if you structure your interactions with Claude.
-
Pete's world: He's been teaching structured interactions from day one (Putting Claude on Rails), but never framed it as an alternative to structured data. His students assumed they needed clean data to start.
The implication: you don't need to organize your files before you can use Claude effectively. You need to organize your approach — project plans, sub-projects, review cycles, git commits.
Key Points
- Structured data → Claude excels (Dave's first two days, building maps and databases)
- Unstructured data + unstructured interaction → chaos (Dave's days 3–5)
- Unstructured data + structured interaction → still productive (Pete's methodology)
- The structure is in the workflow, not the files
Related
- Putting Claude on Rails — the methodology that embodies this principle
- Dave's Claude Struggles — what unstructured interactions look like
- PKAI Course — where Pete will teach this insight going forward
- Handwriting History Project — Dave's case study in messy, unstructured research data