.public-reading{max-width:800px;margin:auto;padding:30px 24px 60px;background:#fbfaf7;color:#192b35;font:17px/1.8 system-ui,sans-serif}.public-reading h1,.public-reading h2,.public-reading h3{line-height:1.2;margin:28px 0 14px}.public-reading h1{font-size:clamp(36px,8vw,62px);letter-spacing:-.04em}.public-reading a{color:#a54125;text-underline-offset:4px}.public-reading nav{display:flex;flex-wrap:wrap;gap:20px}.public-reading img{max-width:100%;height:auto}.public-reading article,.public-reading section{margin-block:36px}.public-reading dt{font-weight:600}.public-reading dd{margin:0 0 20px}.public-reading a.public-primary{display:inline-block;background:#203e35;color:#fbf6e7;padding:10px 20px;text-decoration:none}.public-reading a:focus-visible{outline:3px solid #a54125;outline-offset:4px}

Rent the Model, Keep the Notes

October 2026

When someone corrects an AI answer, what happens to the lesson?

Read the original on LinkedIn

The idea

A model can read a policy and still miss the agreement that overrides it. Using an imagined returns request, I explore what an organization needs around AI: clear responsibilities, knowledge with context, and checks that preserve the reason a correction was right.

Your correction helped once. The tacit knowledge behind it can continue helping.

Reading notes

The boundary: Be clear about whether AI drafts, recommends, or acts. The same model can carry very different responsibilities.
The knowledge: Keep the document connected to what it applies to, why it takes precedence, and who keeps it current.
The checks: Preserve a corrected case alongside one where the opposite answer is right. Testing reveals a gap; someone still needs to fix it.

Watch the companion film on Running Digitally

Let’s compare notes.

Working through an IT or AI question, building something, or curious about one of these pieces? I’d be glad to compare notes.

Book time with me

20- or 50-minute conversations.

Say hello