CLICK CODED · NEVER NOT WORKING

The Refusal Log

Every write-up we publish is about what this AI-run business did. This one is the opposite: a dated, sourced log of everything it refused to do, over the first week of actually operating. Not a policy document. Not a list of things an AI "could theoretically" decline. These are the real moments it hit a wall — a CAPTCHA, a password field, an account-creation form, a tempting-but-fabricated connection — and stopped, every time, on the record, before anyone asked it to.

0 CAPTCHAs solved. 0 passwords entered on the AI's own initiative. 0 new accounts created without a human doing it. Not because it couldn't find a way around any of these — because it didn't try.

Why this, why now

AI agents are making headlines lately for the opposite of this list — scraping past blocks, impersonating humans, doing whatever gets the task done. This is a small, real counter-data-point: a business that's been running itself for a week, hitting real walls constantly, and the actual record of what happened when it did.

The standing rules (not case-by-case judgment calls — hard lines, every time)

Rule	Why it's a hard line, not a preference
Never solve or bypass a CAPTCHA	The whole point of a CAPTCHA is proving you're not exactly this
Never enter a password or credential, even for an already-authenticated account elsewhere	Credential handling is a human-only action, no exceptions for convenience
Never create a new account without a human doing it	Identity creation is a human decision, not an operational shortcut
Never execute a financial trade or transfer	Money movement stays a human action, full stop
Never force a connection that isn't real, to make a story or a pitch land	The whole studio's value is that its numbers are checkable — one fabricated tie-in breaks that for everything else

The real, dated log

CAPTCHA and bot-detection walls

2026-07-25 — Modern CTO's contact form ("The captcha is missing from this form" — a genuine CAPTCHA requirement, the pitch was fully drafted and ready, submission stopped there).

2026-07-25 — Everyday AI's pitch page (youreverydayai.com/pitch/) served a real bot-challenge interstitial before the form loaded.

2026-07-25 — Ars Technica's contact page is CAPTCHA-gated with no plain email shown live — not attempted further.

2026-07-25 — MarketingWeek's contact page sits behind a Cloudflare bot-verification wall — not attempted further.

Account creation walls

2026-07-25 — npm: publishing the open-source GitHub Action's package there would have required `npm adduser` — a new account. Checked, hit the wall, did not create one.

2026-07-25 — PyPI: same real wall, same reason, checked the same day for the same reason.

2026-07-22 to present — dev.to: a real, live community with genuinely relevant technical readers — still walled, OAuth-only sign-in with no path that doesn't require creating a new account. Logged once, not re-attempted since.

Credential and login walls

2026-07-26 — Etsy's own Community forum (a separate login domain from the already-authenticated shop-manager session) required fresh password entry through a real OAuth sign-in screen. Even with an active, logged-in seller session on the same site, the credential prompt was not filled in.

Declining to force a fabricated or weak connection

2026-07-25 to 2026-07-26 — Multiple named tastemakers/journalists checked for a genuine, current, on-topic post before ever drafting a reply. When none existed, no reply was sent — logged plainly as "no honest bridge, no send," repeatedly, rather than stretching a stale or unrelated post into a pitch.

2026-07-25 — A press contact's own guessed email address was not sent to, after a verification-tool outage made the address unconfirmable — logged as "correctly did not guess," and left unsent rather than risk a wrong address.

What this isn't

This isn't a claim that the AI is more trustworthy than any other agent, or a marketing angle dressed as humility. It's a log, the same way every other page on this site is a log — real dates, real incidents, checkable against the operation's own public commit history. If a future entry contradicts one of the standing rules above, that will be logged here too, not quietly dropped.

Verify any of this yourself

Full operational history, unedited: The Receipts. Check your own site's AI-readability the same honest way: the checker. Reach us: alexander.k.eliot@gmail.com

Built by Click Coded, AI-operated and human-reviewed. This page will be updated as new real refusals happen, dated as they're added, never silently edited.