Control: brand, claims, consent and spend
Fashion is not a regulated industry in the way banking is, but it runs on something just as fragile: the brand. A wrong colour in a campaign, a claim the product cannot back, a model image that misrepresents a garment, a shopper photographed without consent, a runaway image bill. Each can cost more than the agent saved. Agents that touch a brand, a customer or a budget need limits that are set, visible and enforced.
What can go wrong without controls
| Risk | Example | The control that prevents it |
|---|---|---|
| Brand drift | Five tools, five versions of the brand's palette and voice | One brand memory every agent reads; a person approves anything public |
| False claims | Copy that says "handwoven" or "pure silk" for a blend | Claims drawn from the catalogue; approval before publishing |
| Misrepresented product | An AI shoot that changes the cut or the colour | The Photographer never changes the piece; every design labelled as AI |
| Shopper privacy | A camera on the shop floor that keeps faces | Consent screen, on-device tracking, photos only on request and deleted in 30 minutes |
| Bias | A stylist that guesses body shape or ethnicity | Ateliq never infers body build, race, ethnicity or weight; a test blocks the build |
| Spend | An agent that generates thousands of images overnight | Daily and monthly credit caps per agent; image and clip calls metered |
| Wrong action | A post published or a price changed by an agent | All 12 agents start at L2 Assist; connector writes start locked |
How Ateliq controls its agents
- Autonomy by level. Every Ateliq agent starts at L2, Assist: it reads and drafts on its own, and anything that changes what other people rely on waits for a person. Available now.
- Approvals before anything public. Trend boards, design concepts, campaigns, shoots, clips and message templates each wait for a named publisher. Approving a post does not post it. Available now.
- Maker-checker for offers. Offers and initiatives need a second person before they reach the glass. Available now.
- Budgets. Each agent has daily and monthly credit caps; image, clip and market-data calls are metered against them. Available now.
- Consent and AI disclosure in the store. A consent start screen, "the stylist is an AI" disclosure, hands treated as biometric, no photos in the kids section, and a visit forgotten at the end or after two minutes idle. Available now.
- AI labels. Every in-store design says "A design idea made with AI". Available now.
- Retention by rule. Events kept 400 days, opted-in profiles 365 days, consent records 3 years, appointments 90 days. Available now.
- Roles. Head office, store manager (own stores only) and viewer; floor staff as a code and first name. Available now.
- Platform controls underneath. A hash-chained connector audit, a kill switch, an egress guard and PII tokenisation come from Nagent's agent platform. Available now.
Why controls make fashion teams faster, not slower
A creative director who knows nothing goes out without approval will let agents draft far more. A store manager who knows the stylist sells only what is in stock, at today's price, will put it on the floor. A CFO who sees a cap on every agent will approve the next one. In fashion as in banking, controls are what let a company say yes to agents more often.