# The dogfooding trap

Argument from [[sail-muddy-lessons]]: dogfooding feels like validation but mostly isn't.

Sail/Muddy migrated off Discord onto their own product, tracked tasks instead of using Linear, reduced Notion to a fallback for when their text editor wasn't enough. The team ran the company on it. Every interaction was smooth because they had authored every concept.

External users walked into the same product and hit a wall: "What am I supposed to do with this?" Too many concepts, too many kinds of spaces, too steep a learning curve. The dogfooding had only proven the product was bug-free and felt good *to people who already knew the model*.

## Why the trap is structural

You can't unknow your own design. If you invented every primitive, the navigation feels obvious. The empty state feels suggestive. The shortcuts make sense. None of that translates to a stranger who has to build the model from scratch.

This is especially severe for tools-for-thought and collaboration products, where the value depends on users adopting an internal mental model. It's less severe for tools that solve a single concrete external problem (file upload, screen recording, payment) where the dogfooding signal is closer to the user signal.

## Slack as a counter-example

Slack famously came out of dogfooding — built as an internal tool during the Glitch game project. But what made Slack land wasn't the dogfooding itself. It was the external signal: when Butterfield opened a preview in August 2013, 8,000 people requested invites in two weeks and 93% of users who tried it never stopped. Dogfooding gave them a polished product. The external signal was what told them they had a market.

The pattern: dogfooding is a quality-assurance tool, not a market-validation tool. Treat the two separately.

## What to do instead

Alejandro's frame: you're often solving interface problems when you should be solving workflow problems. The interface is just a conduit. What actually matters is the workflow and the sensation of completing it — the dopamine hit when you send a message, the feeling of accomplishment when you check something off. Dogfooding can't tell you whether someone *outside the team* wants those workflows in the first place.

## Related

- [[multiplayer-by-default-failure]] — multiplayer products are especially exposed because the team is, by definition, a tight group of collaborators
- [[positioning-vs-vision-gap]] — dogfooding can hide the fact that you can't describe the product
- [[reps-as-pattern-recognition]] — what real external reps look like