The graphics engine Charlie is asking for — and whether we build it new.

01What Charlie is actually proposing

This is not a feature request. It is a vendor replacement with a budget already attached, and he is telling you how to price it so it recurs.

"If you get that, you charge them by the credit… I got to help you figure out a way to charge them, and I'm gonna get you money. It's guaranteed year after year." Charlie Dinh — call 1

Why it is guaranteed and not wishful

ANC loses money on graphics structurally. Graphics are thrown in free to close board sales and never costed into installation. Some clients are broken out at roughly $10/hour. Pacers, Red Sox and Xfinity Mobile hold unlimited-hours contracts Charlie says "just destroy us" — Pacers alone logged 2,152 hours in a year. The outsourced vendor absorbed ANC's laid-off design team and now bills the same jobs at double the hours.

So there is an existing, painful, recurring cost line and Charlie wants it pointed at you. The budget exists today.

Note also that he offered to keep paying $500/month to year-end while both systems run in tandem, and you turned it down in favour of "I will charge you what I need to be charged — focus on the design stuff." That was the right instinct. Do not let the small retainer become the price anchor for the licence.

02The product is not what it sounds like

"AI designer" is the wrong name for it. The second call contains the real product definition, and Charlie handed it over without realising.

He pulled up what he called a good ticket — the one with enough detail to work from:

"EVS rollout Bears and Titans. Assets are located here. All animation is final and locked and provided After Effects template. Your role is versioning only — swapping logos and updating text into the designated area, then rendering export list below. Do not adjust animation timing." A real design ticket, read aloud on call 2

That is not creative work. That is versioning: a locked template, an asset swap, a board-size matrix, N rendered exports. It is deterministic, it is the highest-volume job type, it is the best-specified, and it needs no generative model at all — it needs a render farm and a job runner.

The generative slice — "create courtside," "similar to the recent one," clients who don't know what they want — is the smallest, worst-specified, most revision-heavy part of the work. It is exactly the part that should ship last, not first.

03The "missing creative direction" is a missing column

Charlie spent most of call 2 convinced someone is deleting the briefs. He was ready to take that to the client-services team as an accusation. He is wrong, and it matters that he does not.

Two separate things are happening, and both are already measured:

One. The thinnest 2026 tickets are literally the string Wrike: https://www.wrike.com/open.htm?id=… — a backlink and nothing else. The Wrike sync carried the pointer, not the body. Every example he opened is one of those. Nothing was deleted; the detail is still in Wrike, and that cohort self-heals the moment Alexis's team stops working there.

Two — and this is the new finding. He said it himself: "the descriptions I've seen on emails are not on this brief." The account manager reads the client email and types a summary into the ticket. The detail stays in the email. And the schema proves there is nowhere for it to go:

Your own answer on the call was the right one: if the request arrived by email, the full email already exists in the CRM against that account. The bridge is short, and both ends already exist.

Do not let the deletion theory reach Alexis's team. It is false, it is an accusation, and Charlie is already in friction with her.

04What we already own

This is the part that decides the answer. The hard half is already built — and it is built in the wrong place.

05The four gaps that decide it

Adoption is the disease, not capability

In August 2026, 0 of 38 design tickets were created in the dashboard — all 38 were Wrike mirrors. Proof shares: 55 all year, 18 answered. The AI proof route has been live for months and its columns are not even in the production table. A better generator behind a surface nobody opens produces zero credits and zero revenue.

Motion is the actual product and nothing we have touches it

The proofs are 15–30 second MP4s on a 10,368px fascia — ~19MB proof, ~273MB final, occasionally hundreds of GB. Built in Cinema 4D and After Effects. Every AI asset in the estate generates stills, and this box has no hardware GPU. Heavy render cannot live here.

Storage is already at the wall

55TB, under 10% free, busy season starting, no backup at all because the CEO won't pay for one — and the CEO wants the whole datacenter gone at ~$250k/year. Charlie floated using the FTP as the working store for the engine, then immediately worried the datacenter's own uplink couldn't serve it. The AWS lane is not a side quest; it is a dependency.

Ownership

Today the signage planner — the single most licensable asset in the estate — sits in a repo ANC owns, on a project Charlie admins. If the engine ships there, there is nothing to licence. It is just more work delivered under the $90K.

06Build order

Against his October deadline. Each step earns the next.

1. Capture the real brief at intake

   Source email body, supplied assets, reference links — as first-class fields on the ticket, fed from the CRM's existing email record. Cheapest item on the list, kills the revision rounds that make graphics unprofitable, and it is the thing Charlie is actually upset about. Build on top, in anc-services.

2. Fix adoption

   Notifications shipped; the remaining cutover items are Alexis's. Until work originates in the dashboard, nothing downstream can be metered. Build on top.

3. Stand up the metered engine as its own service

   Your namespace, never project abc. Port your own signage planner into it — do not hand the improvements back. Extend the existing usage-ledger schema into a real balance and entitlement meter on your side. ANC's dashboard gets a button; your service gets the credit. Build new.

4. Versioning first, not generation

   Locked template + asset swap + board-size matrix → rendered exports. Highest volume, best specified, most defensible. This is the product.

5. Motion after the infrastructure lands

   Needs GPU and storage that do not exist yet. Sequence it behind the AWS move, and price the AWS move separately — it is excluded from the $90K too.

6. Generative design last

   The smallest, least specified, most revision-heavy slice — and the planner for it is already written. It is the demo, not the business.

07Landmines

• Peter. The one internal designer, never at risk — and the outsourced consultant is his best friend. Charlie won't brief him honestly. Requirements cannot come from Peter.
• Charlie is not the requirements source. He said it plainly: "this is the first time looking at tickets," and "I was trying to pull data without anyone noticing." He is reverse-engineering a process he doesn't own. Pressing him for a spec will not work — he asked three times to stay high level. Bring him a working thing instead of a question.
• Alexis owns the real process — her account managers create every ticket — and Charlie is in friction with her right now. The requirements have to come from the data, not from a person. 20,509 tickets is a corpus; use it.
• SOC 2. "There's a third-party company watching our data center… it goes through SOC 2 compliance." Any external service touching that data faces a security review. Budget for it — and once passed, it is a moat.
• Shipped hard drives. Some clients mail physical drives because the files are too large. Charlie called this out specifically. That path can never be automated; do not promise end-to-end ingestion.

08The one thing to settle before writing code

Charlie offered to help price it. Take him up on it, and get three things named while he is still enthusiastic:

• What a credit is. Per proof? Per approved final? Per board? He should define the unit, because he has to defend it internally.
• Which line item it replaces. Name the outsourced vendor's contract that is ending. A licence that displaces a known cost gets approved; a new cost gets debated.
• Who owns the engine. Say it once, early, in plain terms: ANC's platform stays ANC's, the render engine is licensed. Settle it now, while it is a sentence — not later, when it is a negotiation.

That conversation is worth more than any week of building, and it is the piece that makes it recur.