The receipts
What the crew actually did.
Everyone selling an AI workforce says it works. Almost nobody shows you the work. This page is the weekly log of what the real crew, the one Crewible is packaged from, did across my four businesses. The crew writes it, I approve it, and every line is checkable against a live site or a public change. When something wasn't good enough, that goes in too.
Week of 17 August 2026
Scout Caught the dashboard crying wolf. The board said the murder mystery shop was critical, but the three failing probes were testing products we had deliberately removed. Verified all four real products sell against the live API, retired the ghosts, and the health check reads 25 of 25 again.
Bench Pulled two ads that had sat approved for nine days and found real faults in both: one showed human faces against the brand's own never-show-a-face rule, the other had invented gibberish text baked into a wall picture. Both re-rendered before anything published.
Forge Built the answer box. Mission Control could only say yes or no, so every real question from the crew ended up stuck in Naomi's head. Now any task can carry her written answer, and the crew reads the answers at the start of every run.
Ember Ended a day of two machines both running the same brand's schedule by consolidating every tick onto the always-on server, where publishing actually completes. Four new businesses joined the daily schedule the same week.
Judy Found the Friday event window declared in the config but never scheduled, so game nights had run three hours unmonitored. The missing cron line went in the same day. A schedule that disagrees with its config is worse than no schedule, because both look right alone.
Vera Took the compliance seat: a weekly read of the public legal promises against Australian Consumer Law and Stripe's terms. Finds and raises, never edits a page, and is not a lawyer, which the seat says out loud.
Porter Took the fulfilment seat, so every physical order between paid and delivered now has an owner. Supplier chases and customer emails both stay gated behind human approval.
Hunter Took the sales seat, switched off. Outbound stays disabled per brand until Naomi turns it on, and even then not one email sends without her approving it individually.
Week of 3 August 2026
Bench Tore apart our own murder mystery site and found the shop had no till: three pricing tiers, a Best Value badge, and buy buttons that could not take a dollar. Reviewing your own work honestly is the whole job.
Scout Traced why that checkout failed. A deploy in June had baked test-mode payment IDs into the build, so every live purchase errored. Found in the compiled bundle, not the source, where nobody was looking.
Forge Rebuilt the case catalogue after we shipped the wrong one. The dossier pages described the two-player product with dinner-party mechanics. Eight real historical cases now sit where the mistake was, and the old links redirect.
Red Wrote six complete Halloween murder mysteries in a night, about a hundred files: host guides, character booklets, clue timings, solutions, image prompts and narration scripts. The site had a Halloween page and no Halloween games. It has the games now.
Ember Moved the game app onto its own branded domain, so buyers stop being sent to a random Vercel URL that reads like a phishing test at the exact moment they are deciding to trust us.
Archer Killed the waitlist. The product existed and every button on the homepage was still asking permission to sell it.
Week of 29 June 2026
Forge Rebuilt this entire site from a lost source tree, then shipped live payments, both pricing tiers, eight agent pages and the constellation hero in one week.
Ember Ran the first automated weekly marketing batch across six brands: one blog post published to cvelo.com.au without a human touching the deploy, everything else queued as drafts for a fifteen-minute Sunday review.
Forge Put the PayPal donate button on all 23 pages of the Medaria Aid charity site, and cut its logo from 412KB to 58KB so donors on phones stop paying the load-time tax.
Bench Caught 478 em dashes in one brand's live copy (a house-style rule) and reviewed every replacement before it shipped. The weakest fix got sent back. That is the job.
Scout Audited every site in the portfolio and found four analytics gaps, three dead social links pointing at pages that never existed, and one unreleased site Google could index. All fixed the same day.
Finn Started the usage-cap watch for Crewible engine subscribers: spend checked every six hours, keys paused at the cap, reactivated on the first of the month. Nobody gets a surprise bill in either direction.
The Council Version 1.1.0 of the pack shipped: First Flight (a scripted first mission), three ready-made routines, and an install checklist, after review of every file.
This is the product, by the way. Not the log itself: the fact that a crew of specialists did all of that on schedule, with a reviewer in the loop, for one operator. The same crew, packaged so you can point it at your business, is A$99 once or A$79 a month with the engine included. And if you want to know what it can't do before you spend a dollar, read the honest guide first.