Everything here is running, and we are the first client of all of it.

Turns an approved editorial theme into a published article: an idea queue, drafting from the client’s own brief and facts, a review gate before anything is published, a calendar the client controls, and a results page. Runs unattended on a schedule once the client has switched publishing to automatic.

Twenty articles live for one client, publishing every five days; twenty eight for ProjxAI’s own site. The first sixteen client articles cost A$4.29 in total to draft.

What a client sees at app.projxai.com.au: their home, content library, calendar, publishing controls and results, on one address per client. Every screen is a projection of the record with an allowlist; nothing renders that was not permitted to.

One client workspace in daily use. Editing, choosing a photograph, showcasing an article and requesting a change all travel through audited functions, not direct writes.

What ProjxAI sees at core.projxai.com.au: every client, the queue, review, publishing settings, theme governance and the run record, in one dark room. A client may propose a theme; only ProjxAI may approve one, and that boundary is enforced in the database.

Every client decision since 22 August has a record: who, what, when. Every schema change arrives as a reviewed migration, and the repository replays all of them on a clean database as a standing gate.

An accountability record for machines where AI touches real work: what the AI did, on whose instruction, to whose data, and where it went, on a machine configured so the answer exists. Three layers: the Recorder, the Watcher and the Ledger. The artefact is a one page monthly attestation an independent verifier can check without the product present.

Fifty five configuration checks that took its own author’s machine from twenty one failing to zero. A hash chained ledger that four rounds of adversarial review could not alter silently; every finding is a permanent test. 444 tests passing.

The governed decision process behind the Ideas Lab: success criteria, budget cap and kill criteria signed before work starts; a panel of competing frontier models under opposing role contracts; adversarial review with dissent recorded; a person accountable for GO, FIX or STOP; a dated audit trail.

Its first verdict was STOP, to its own founder, on his own project. Its first client run returned FIX with a five seat panel split three to two, and the client took the advice: validate before building.

Fourteen questions about how a business actually runs, then a personalised report: readiness score with the breakdown, three named opportunities ranked by impact and effort, a ninety day starting plan and investment estimates sized to the business. Written from the answers, not a template.

Reports generate in about two minutes and arrive by email with a permanent link. The repair itself is a small example of the method: it was broken for three months, measured, fixed, and proven with a report before it was called done.