# The placeholder fit on the slide. The real client name didn't. — retained record

> Retained reddit record from reddit:AI_Agents.

Canonical: https://highsignal.app/data/records/54bfed2af4d74212

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:AI_Agents)  published 2026-10-27 03:20 ingested 2026-10-28 02:35

[original source ↗](https://www.reddit.com/r/AI_Agents/comments/1xghsf1/the_placeholder_fit_on_the_slide_the_real_client/)

## retained text



```
One of the awkward problems in building OpenGamma wasn't getting Claude to write a slide. It was getting the slide to survive putting the real information back.

The workflow replaces sensitive values with placeholders before the AI works on the deck, then restores the values in the exported PowerPoint. That creates a small layout trap: \[CLIENT\_1\] is short. "Northwind Logistics Holdings" isn't.

A slide can look fine while it's being designed around the token, then break when the full name returns. The AI has made room for the thing it can see, not the thing the reader will get.

I had to size text for the real value rather than the placeholder. Otherwise, hiding the information just moved the problem to export.

The way I'd explain the design constraint now: don't hide a value and accidentally hide everything the next step needs to work. A layout doesn't need to know the client's name, but it does need enough room for it. The same goes for a table cell or a chart label.

That leaves two separate checks: did the sensitive value stay out of what the model saw, and does the finished deck still work once it's restored? Passing the first doesn't tell you the second passed.

I built OpenGamma with Claude Code. This is one of the less glamorous parts of that build, not a claim that masking alone makes a document safe to share.
```



document fetched 2026-10-28 00:20 · canonical https://reddit.com/r/AI_Agents/comments/1xghsf1/the_placeholder_fit_on_the_slide_the_real_client
