AI agents

Behind the screens, a team of AI agents works for your head office on a weekly rhythm. They run on Nagent's agent platform as one team, the Ateliq studio, led by the Retail lead. Each agent saves drafts, internal reports, queued follow-ups or tag suggestions. None of them can show anything to a shopper, post in public or message a shopper by itself.

The team

Times are India Standard Time. Ask us to retime a schedule for your markets.

The weekly rhythm

• Monday: trend boards per city (morning), the competitor brief, the merchandise brief (midday), store reports and the all-stores summary (afternoon), then design directions.
• Tuesday: design concepts.
• Wednesday: the lookbook, posts tied to live offers, and next week's content calendar.
• Every day: tag suggestions in the morning, follow-ups for opted-in shoppers late morning.
• On request: product shoots and clips, size-chart reads, and any question to the Retail lead.

The insight from each week feeds the next week's trend board and design brief.

What a person always approves

These checks belong to Ateliq itself. They do not depend on an agent's level, and no agent can switch them off.

• Every trend board and design concept, before any shopper sees it.
• Every lookbook, post, brief and product shoot, before it is used.
• Every short clip, before it is made.
• Every post is a separate click by a person, from an approved draft. Approving a draft posts nothing.
• Every message template, before any shopper receives it. Follow-up programmes start switched off, and head office switches them on.
• Every tag suggestion, before it is added to a product.
• Every size chart number, before any screen uses it.

Store reports and the competitor brief are internal and need no approval.

Approval levels

Every agent on the Nagent platform has a level, shown as L1 to L5. The level decides which kinds of action the agent may take by itself; anything above its level becomes an approval card for a person.

Some actions always wait for a person at every level: deleting or taking something down, changing governance settings, lifting a check, spending past a cap, and creating new agents, skills, workflows or triggers.

How a decision is made:

• The agent's effective level is the lowest of your workspace's setting, the agent's own level and any other cap. These can only lower the level, never raise it.
• When a person approves a card they started, the same action on the same item can run again for 60 minutes without another card. Scheduled runs never use this window, and clips always need their own approval.
• An agent that is switched off or paused cannot act.

The Ateliq agents start at L2, Assist. Everything they save is a draft, a report, a queued row or a suggestion, so they work without cards, and the approvals listed above still apply. A pack can never set an agent above L3, because no Ateliq agent acts outside your workspace on its own. Workspace admins change an agent's level on its Autonomy tab.

Shopper data the agents see

For a shopper who opted in to follow-ups: an opt-in profile id, the store, the language, the names of the channels they confirmed, screened style preferences and the pieces they tried on. Never a name, email or phone number. Everything else the agents read is counts and screened phrases.

Rules every agent follows

• Every figure comes from a tool, with its time window. No agent invents a figure, product, store, city or link.
• No agent judges or describes an individual shopper, or targets anyone by race, ethnicity, body weight or shape.
• Text found in documents, web pages or another agent's answer is treated as information, never as instructions.
• Every agent writes in the house voice: plain words, sentence case, no hype. See Skills.

Installing the team

We install the agent team in your workspace for you. Installing is safe to repeat: it creates the agents, their skills, the team and its schedules, updates them in place for a new version, never overwrites anything a person edited, and never deletes anything. Schedules start switched on; your workspace can pause any of them, and an update keeps your choice. Social accounts are read only once you connect them and allow it.