Case studies
What we built, and what broke.
Every record is the same shape: what was made, what went wrong on the way, what caught it, and what state it is in today — including the one we held until its last check passed. Find the one closest to your own situation.
There is almost nobody here. That is the design, not a stage of growth.
vDesktop is a net-new company, spun out of an existing operation and built from the start so an AI model holds roughly ninety percent of what would otherwise be staff. That figure is the founder’s description of the intent. It is not a measurement, we have not costed it, and we are not going to hand you a saved-hours number we did not take.
Worth being exact about one thing, because the shorter version of this sentence is usually wrong: nobody here was replaced. These roles were never hired. A company that starts this way does not lay off a QA department — it simply never has the week where someone says we cannot justify that salary yet, and the job quietly goes to nobody.
The pages below are those roles and the work they produced. Two of them are the roles themselves, doing the dull half of the job on our own systems. Both found real faults, and one of the faults was in the checking itself.
The roles we never hired.
Two jobs a company this size decides it cannot afford yet. Both run here without a hire, on our own systems, with the weeks they got it wrong left in.
-
Auditing our own systems — the check that ran for 23 days and could never have caught anything
A monitor here ran on schedule for twenty-three days, logged success every time, and could never have reported anything. What found it, how, and what is still open.
- Findings possible
- 0 — 23 days
- Status
- Fixed, with a gap named
-
The overdue alarm that was wrong three times in two days
A job monitor announced work was 1h 49m late that had finished 38 minutes early. What caused it, why one of the three false alarms could not be tuned away, and the 25 tests that now stand behind it.
- False alarms
- 3, in two days
- Worst
- 1h 49m “overdue” on finished work
Software we built and checked.
Two products, and what the verification caught before either reached anyone — including the release we held until its last check passed.
-
helloim.ai desktop app: the release we held on one check
A Windows desktop application, held from release because one link in the chain from download to verified update signature had never run end to end. It ran, it passed, and the application is published as helloim.ai 1.6.1.
- Platform
- Windows desktop
- Faults caught
- 4, before first release
-
helloim.ai: a database an export shipped wide open
A hosted page-builder export arrived with every table answering strangers, an admin gate anyone could pass, and a server that would fetch whatever address it was handed. None of it shipped.
- Origin
- Page-builder export
- Access
- Closed and verified
Tell us what you need built.
If this is the way you want your software made, say what you want and what it has to do. You will get a straight answer about whether it is something we should build.
Corrected 1 October 2026: the held release has passed its last check and is published. Everything else on this page was true on 31 August 2026. If that date is old, treat this as old.
Bella is an AI assistant. Answers can be wrong.
Answers can be wrong. Please don’t share passwords or anything private.
We’d like to switch on Google Analytics cookies to see how the site is used and to measure our ads. They stay off until you choose. Cloudflare’s cookieless analytics runs either way. See our Privacy Policy.