I stopped copy-pasting between tabs for 30 days. Here's what the AI assistant that replaced it taught me.
Two weeks ago I shipped a Chrome extension that does the thing I used to do manually a hundred times a day: read the page I'm on, act on it, and write replies right inside the text box. No switching to a chat tab, no copying content back and forth, no context lost in the transfer.
The rule I gave myself: if it adds friction, it doesn't ship. That's how I ended up with an assistant that has a sparkle button next to every input on every site — because the place you type is the only interface that matters.
But building it surfaced three things I didn't expect:
-
The privacy ceiling is the real moat. Every assistant I tried wanted my data more than it wanted to help me. So I built SideKik with zero data retention as the default: prompts are sent, answered, discarded. No storage, no logs, no training. That constraint removes the trust objection entirely — and it's the hardest thing to copy because it means giving up features (sync, history, analytics) most companies won't forfeit.
-
Acting on the page is where the magic is — and where the pain is. "Fill this form with my details" works only if you handle contenteditable divs, shadow DOM, React inputs that reject programmatic values, and sites that re-render mid-keystroke. The chat part is commodity. The action part is a minefield. Users notice the difference instantly.
-
Token metering beats request metering. My cost scales with the size of the prompt, not the count. Free tier: 50K tokens/month. Pro: $9 for 5M. Meter what you actually pay for.
Numbers so far: the wait for Chrome Web Store review is real (submitted, a few days out), but the extension is live for anyone who wants to try it: https://sidekik.win
What I'm most curious about: for those of you who've shipped browser extensions — what did your first 100 installs teach you that you wish you'd known before launch?
The biggest lesson I’d expect from the first 100 installs is that the features you think are the moat may not be what users actually value. I’d watch where people repeatedly use the extension, where they drop off, and what they try to do that you didn’t anticipate. Those patterns can tell you what deserves to become the core product.
"The place you type is the place you need help" is a sharp insight. Zero retention by default is a real differentiator; I'd put it front and centre on the listing, since that's the first objection people have about an assistant that reads every page.
The split between “chat” and “action” is a useful product boundary. I’d instrument failures by DOM shape (contenteditable, shadow DOM, controlled input, re-render) and preserve the smallest reproducible page state, because aggregate success rates will hide which integrations are brittle. For the first 100 installs, segment those failures by site and workflow rather than only counting installs; that should tell you which compatibility work is worth doing.
This is great work — reminds me of some of the calls I've had to make building Xstream4K. What would you do differently if you started over?