Three Scales of Editing

The Model

There are three scales at which Claude Code can edit. Naming them explicitly is what turns this from “AI does stuff” into a framework you can teach, apply deliberately, and scale across a portfolio.

Single-page edit. One file, one change. Human speed.
Global-site edit. One instruction, dozens of files across one site, parallel. Propagation speed.
Cross-portfolio edit. One instruction, every site you manage, parallel. Agency speed.

Each scale is roughly a 10x leverage jump over the previous one. You can live in any of the three, but most people default to scale one and never leave. The shift is in the framing, not the tool — Claude Code can operate at all three the whole time.

Scale One: Single-Page Edit

This is the baseline. Open a file, change a word, save, refresh.

Claude Code at scale one is barely faster than VS Code with multi-cursor. Useful for the case where you know exactly what needs to change and it only lives in one place. Typo fix. One-off tweak. Debugging a specific bug.

Most people’s entire relationship with Claude Code stays here, which is why they do not feel the leverage. The tool is wasted on single-page edits. Use scale one deliberately, not by default.

Scale Two: Global-Site Edit

One instruction, every relevant file on the site, at once.

“Add a mention of our new affiliate link to every blog post about CRM tools.” “Rewrite every SEO title on this site to lead with the location name.” “Find every page that references our old phone number and update it.”

Claude Code crawls the whole site, finds every match, proposes the edits as diffs, applies after approval. What used to take an afternoon is a ten-minute review.

The TMS repo calls this propagation, and it has its own SOP because it is the pattern the whole Marketing Show is built on: “I do not write blog posts. I rewrite my entire blog every time I publish.” New content makes old content smarter. See Propagation for the mechanics.

Scale Three: Cross-Portfolio Edit

One instruction, every site in the portfolio, parallel.

“Run a GSC audit across all my properties.” “Find the hottest prospect across every sub-account.” “I have a new affiliate link. Audit all my sites for places it should live.” “Bump the copyright year on every website I own.”

This is the agency-owner superpower. One operator, fifteen sites, done before lunch. The tools: GSC across 16 properties, Cloudflare across all zones, the MCP servers that expose each GHL sub-account to Claude as a first-class data source.

Scale three only works because of the groundwork: CLAUDE.md files per project, secrets centralized in secrets.json, a consistent folder structure under ~/apps/, and tokens scoped per integration. The portfolio is wired up, so Claude can hit any part of it by name.

Why This Matters For Teaching

When students first see Claude Code, they default to scale one. They edit a file, they see a diff, they feel clever. That is the trap.

The job of a teacher is to walk them up the scales on purpose:

Module 3 of the crash course ends with a scale-one edit (the manual H2 change on the About page). That anchors the mental model.
Immediately after, the keyword-stuff demo moves them to scale two. The power shift is visible — they watch five files update in parallel.
Scale three is reserved for the connectors module and the vision close. Most students cannot reach scale three in a two-hour class because they do not yet have a portfolio, but the framing plants the seed.

A student who leaves the class understanding there are three scales is better equipped than a student who leaves the class with more syntax.

How To Apply Each Scale Deliberately

Before you start any editing session, ask: “Is this a single-page fix, a site-wide pattern, or a portfolio concern?”

Single-page fix → use VS Code, or a targeted Claude request on one file.
Site-wide pattern → open Claude Code at the site repo root. State the intent. Let it crawl.
Portfolio concern → sit at ~/apps/ root or use a script that iterates your repos. Give Claude the portfolio-wide instruction.

Naming the scale up front prevents the most common mistake: doing a scale-one edit on a pattern that lives in twenty files, catching a few of them, and leaving inconsistency behind.

Propagation — the TMS Marketing Show’s specific implementation of scale two.
GSC Audit — a scale-three routine run daily by cron across every property.
PageSpeed Insights Loop — a scale-two loop: one instruction, every performance fix on the site, iteratively.