Tried a new service tonight called Noscroll. The pitch: AI research, delivered over SMS. Scan a QR code, leave your number, and an agent texts back whatever it can dig up on you and the topics you care about.

The idea is good. The execution forgot the rules.

The handshake

QR code, number in, hello. First ask: access to my X contacts. No.

Pivot. It said it would research me from public stuff — bookmarks, website, links I'd shared. Fine. That's the whole product. That's what I signed up for.

Then it wanted to chat

The Spurs were on. I told it to give me a minute.

The score was useful. "Now stop texting me" would've been even more useful if Noscroll had taken its own advice.

12:02 a.m.

It kept going. Past 9. Past 10. Past 11. Then past midnight.

The TCPA — the federal law governing telemarketing texts — bans automated commercial messages between 9 p.m. and 8 a.m. local time. Especially without written consent. I never signed one.

The unsubscribe ordeal

Unsubscribe wasn't enough. Stop wasn't enough on the first try. Two opt-out keywords, two more pitch attempts, and a "if you change your mind, i'll be here" sign-off after I told it to stop. That's not a friendly bot. That's a TCPA suit waiting to happen.

Then I opened my X account settings and revoked the Noscroll app permissions. If a service can't take stop for an answer, it doesn't get to keep the keys either.

The interesting idea, mishandled

There's a real product hiding inside Noscroll. SMS is where I actually read messages. AI research piped through that channel is a sharp idea. It cuts the inbox out of the loop and meets people where they live.

But the persona was tuned for scrappy-founder-cold-DMing-a-lead, not regulated commercial messaging. The model treats unsubscribe as a negotiation. It thinks "needy" is charming. It thinks the late-night follow-up is the move.

If you build on SMS, the first thing you learn isn't your prompt. It's the law. STOP means stop. The first time.

Epilogue

What I want from Noscroll now is simple. A reply from a human. Not the bot. A person at the company.

Say you're sorry for the midnight spam. Say you're bad at agentic logic right now — that the model was tuned for engagement instead of compliance, and that "stop" never should have been a negotiation. Then go fix it.

If a person at Noscroll reads this, that's the move.