I thought users would prefer the extension. They trusted the web app more.

I had an unexpected realization while building Lakon.

I originally thought the browser extension would be the main value:
one-click prompt compression directly inside ChatGPT/Claude.

But after talking to people, almost everyone preferred the web app instead.

Not because it was more convenient, actually the opposite.
It adds an extra copy-paste step.

The real reason was trust.

People wanted to manually verify what changed before using the prompt. A lot of users said extensions feel invasive now, especially when they touch AI conversations.

So I stopped treating the extension as the “main product” and started focusing on making the web experience genuinely useful:

• preserving intent
• reducing token/context waste
• showing exactly what changed
• making outputs easier to trust

Honestly, this feedback probably saved me months of building in the wrong direction.

Curious if other builders here have had moments where users behaved completely differently from what you expected.

This is what I am building
-> Lakon