The job is to replace ANC's outsourced design vendor with an AI pipeline by late October. Everything here is either verified against a live system or flagged as unverified.
There are three design programmes, not one. Conflating them is why the asks feel endless.
| Programme | Owner | State |
|---|---|---|
| Services Design | Alexis + Joe | Built, adoption collapsed |
| Sales-support Design | Daniel Croci via Natalia | Parked — political gate |
| AI Designer | Charlie | The actual goal |
Two separate retirements bundled into one scope doc, which is why they blur. Wrike is the old project tracker → replaced by the Service Dashboard design module; all the notification and brief work lives there. Airtable is Nick's ops bases, the other half of that same doc. NocoDB is the operations dashboard — a separate product that was never the Wrike replacement.
August design tickets created in the dashboard. All 38 were mirrors from the old tracker. Monthly dashboard-originated: 27 in April, then 1, 5, 18, and now zero.
| Adopted | Built but dead | ||
|---|---|---|---|
| Designer time entries | 23,988 | Proof shares | 55 |
| Content schedules | 3,036 | Print requests | 21 |
| Design requests (mirrors) | 20,506 | Recurring-client templates | 1 |
| CG requests | 633 | Ticket files / attachments | 9 / 0 |
Everything that helped someone do their existing job faster got adopted. Everything that asked them to change where they work did not. That is a switching-cost problem, and the switching cost was notifications — now fixed.
Client emails a request → account managers create the ticket and move assets to the storage server by hand → Alexis assigns a designer → a low-quality proof goes to the client → approve or deny with a reason → only then is the full-quality file rendered, which can run to hundreds of megabytes.
The storage holding every proof and finished asset is about 55TB, under 10% free going into busy season, with no backup because the spend was declined — and leadership wants the data centre gone at roughly a quarter million a year. Anything built on top inherits that. Raise it in writing before the AI designer work starts.
| # | What | Why it mattered |
|---|---|---|
| 1 | Jireh's Microsoft sign-in | Redirected to localhost, so he never landed. Hit every SSO user. |
| 2 | Status changes by email | Was chat-only and silently skipped 99 of 189 staff. |
| 3 | Notification reliability | Delivery log, health page, daily watchdog. It can't fail quietly now. |
| 4 | Brief gating at intake | Kills the vague briefs that drive revision rounds. |
| 5 | CRM search ranking | Exact-match opportunities were buried under weak-match accounts. |
Also published: design.basheer.app (programme map) and requests.basheer.app (master tracker — bind-mounted, edit the file and it is live).
A fresh last_login_at does not prove someone got in — auth
succeeded one line before the broken redirect, so Jireh's account looked perfectly healthy while
he could not reach a page.
On the CRM search: inverting the account-first rule is wrong. Ranked purely by relevance the Rockets account falls to #14. The fix was a full-match tier above that rule, not a replacement for it.
The actual goal. Brief plus assets in, proof out, into the approval flow that already exists. The pieces are readier than they look — proof server connected, approval workflow live, image generation already wired. Needs to work by late October. Billed per credit.
New and untouched. The renderer lives in the proposal engine, not the CRM. Two example files were attached to her message.
Duplicate a request, templates per recurring client, venue-tree sidebar, day-before reminders. Templates already ship and hold exactly one record.
A link, not copied data.
Takes the chasing off Nick and Gianni.
Feeds the estimator. Wants a backfill from 2023.
Ahmad owes this. Use questions that reconcile to SQL so a wrong number is provable.
Ahmad owes both. Biweekly Wednesday, Natalia included.
Ahmad owes the invite. She is open to it and has not seen it.
Ad Campaigns, Black Friday, Summer Sale. Some may be live — remove carefully.
His cadence, two to three months.
Jireh + Alexis + Kirsten have to agree it first.
Raise in writing before building on it.
~6,700 records, a field-type change.
242 created in one 20-minute run. Bulk destructive — needs a go-ahead.
The first is gated by Natalia, the second paused by Jireh, the third sits behind design sign-off by Alexis's own sequencing.