How Can an Agency Run AI Agents Across Many Client Stores?
An agency runs AI agents across many client stores by keeping one shared playbook and separate access for each client. Each store gets its own Noibu domain, teammate permissions and repository connection. Agents find issues and draft changes store by store, a named agency reviewer merges each one, and the client approves what goes live.
TL;DR
- Keep one shared playbook and separate access for every client store.
- Each client gets its own Noibu domain, teammate permissions and repo connection.
- Agents find and draft; a named agency reviewer merges; the client approves what goes live.
- Report each client's results in orders and revenue, not hours.
Agencies that serve many merchants are asking how to run AI agents across every client without mixing data, access or approvals. This playbook covers the operating model, the access setup for each client store, what to schedule and how to report. For the bigger picture of the model, start with what an AI-native ecommerce agency is.
What does running AI agents across client stores involve?
Running AI agents across client stores means two layers. The shared layer is the agency's playbook: the skills and checks agents run, the rules for reviewing their output, the approval template and the reporting format. The per-client layer is everything that belongs to one merchant: its store data, its Noibu domain, its repository and the people allowed to approve changes.
The shared layer is what makes the model scale. John Duncan, CEO of the Salesforce Commerce Cloud agency 64labs, describes this in Managed Agents, Not Managed Services (August 31, 2026): one person can manage shared agent infrastructure across a whole client base, and a lesson learned on one client's build ships to every client's build. The same article sets the rule that keeps it safe: a named human reviews every merge.
The per-client layer is what keeps the model trustworthy. Each merchant's data, access and approvals stay in their own lane, so an agent working on one store never sees another client's sessions or opens a pull request in the wrong repository.
One shared playbook, separate lanes per client
Shared across all clients
Skills and checks, review rules, approval template, reporting format. A lesson learned on one store improves the playbook for every store.
Client A
Client B
Client C
Data, access and approvals never cross between lanes.
How should an agency set up access for each client store?
Access in Noibu is set per account and per domain. A client adds agency staff as teammates from the Settings menu in the Noibu console, and Noibu does not limit the number of users an account can add. Each teammate gets a permission level: Full Access covers issues, session replay, performance and configuration, while Sessions Only covers the Sessions and Live View pages for selected domains. At least one teammate on the account must keep full access, so the client stays in control of its own account. The steps are in Noibu's help articles on adding and removing teammates and permissions.
Clients with several storefronts, such as regional or brand sites, can keep them under one Noibu account. A Noibu account can be associated with multiple domains, users switch between them from the dropdown in the top-left corner of the console, and each domain has its own settings. Adding a domain means deploying the Noibu script in the site's head tag and asking the Customer Success Manager to associate it, as Noibu's help article on multiple domains explains.
Repository access follows the deploy path. When the agency holds the client's code, agents open pull requests in the agency-controlled repository and the agency's developers review them like any other change. The setup for that case is in how to connect AI agents to GitHub when you don't own the repo. In one agency engagement, an agency that holds a retailer's Bitbucket repository reviews and merges agent pull requests through its own ticketing system, with no production access given to the agents.
How does an agency keep each client's data and approvals separate?
Separation comes from setting access per client rather than sharing one login across all of them. Agency staff are added to each client's account as named teammates, scoped by permission level and, for Sessions Only users, by domain. When someone leaves the agency, they are removed from every client account they were on.
Approvals are separated the same way. Each client has a named reviewer at the agency and a named approver on the merchant side. Agents draft the change, the reviewer previews, edits or rejects it, and the merchant signs off on what reaches the live store. To stop pull requests piling up across many clients, use the batching and ranking rules in how to keep AI-generated pull requests reviewable.
What should run on a schedule and what should run on request?
Scheduled runs are where agents earn their keep across many stores, because problems surface between client meetings instead of waiting for a ticket. That matters because over 90% of website errors are never reported by shoppers. A daily brief is the usual starting point; Noibu's Store Pulse skill can be scheduled as a recurring brief to Slack or email, and a daily AI report on errors, speed and conversion shows what it covers. Whether a run can be scheduled depends on the AI tool's own feature support.
Agree the schedule with each client before turning it on. Some agencies bill only for work a client requests, and in one engagement an agency on Noibu MCP chose not to run scheduled agents from its own account for that reason. Writing down which checks run daily, which run per release and which run only on request keeps the retainer and the agent workload aligned.
How should an agency report results across many clients?
Report each client on its own numbers: what shipped, what it changed in conversion, speed or revenue, and what is queued next. Use the same format for every client so the agency can compare stores and spot which playbook changes are paying off. The method for measuring a single change is in how to know if a website fix actually improved conversion.
The multi-client setup checklist
Use these six steps for every client store an agency brings onto AI agents:
- Confirm who owns the deploy path. For each client, write down who holds the repository and who merges. If the agency merges, agent pull requests land in the agency's review queue; if the client merges, the agency reviews and hands over.
- Get teammate access to each client's Noibu account. The client adds agency staff from Settings in the Noibu console. Give reviewers Full Access and limit anyone who only needs session replay to Sessions Only on the right domains.
- Set up every storefront as a domain. A Noibu account can hold several domains, such as regional or brand storefronts. Settings like issue labels and the Jira integration are configured per domain, so set them for each one.
- Connect the AI tool and the repo per client. Connect Noibu to the AI tool the agency uses, such as Claude, ChatGPT, Microsoft Copilot or OpenAI Codex, and connect each client's GitHub repository so agents can open pull requests there.
- Agree the approval rule in writing. Name the agency reviewer for each client and what the client signs off on. Nothing ships without approval, and agents never get production access.
- Report per client in orders and revenue. Send each client what shipped and what it changed in conversion, speed or revenue, measured on that store's own traffic.
Where do Noibu's agents fit for agencies?
Noibu gives an agency the per-store data and the agents in one place, in the Noibu console and through Noibu MCP in AI tools teams already use, such as Claude, ChatGPT, Microsoft Copilot, GitHub Copilot and OpenAI Codex. The Bug Resolution Agent finds the top fixable errors on each client store, traces root causes in the codebase and opens a pull request for review. The CRO Agent finds underperforming product pages and rewrites them, and on Shopify inline edits publish in one click once approved. The Performance, ADA, A/B Testing and ROAS agents cover Core Web Vitals, accessibility audits against WCAG, experiments and paid traffic scored on orders, as described on the AI agents page.
Every agent runs the same loop: find, build, approve, ship, measure. Nothing ships without approval, so the agency stays the reviewer and merger of record and the merchant keeps the final say. Agencies use that to do more for each client; Montreal agency Blue Badger describes the commercial side in its customer story, reporting a 20% increase in client retainers after adding Noibu.
Noibu is AI for ecommerce operations, the always-on optimization system that keeps online stores running at their best. A team of supervised AI agents finds what's broken, slow, or leaking revenue, does the work, and hands results back for human approval. Trusted by retailers like Free People, Guess, Liquid I.V., and Ariat.
Frequently asked questions
Can one agency team run AI agents on many client stores at once?
Yes. The agency keeps one shared playbook of skills, checks and review rules, and connects each client store separately. Agents work on each store's own data, and a named reviewer at the agency merges every change for that client.
How does an agency get access to a client's Noibu account?
The client adds agency staff as teammates from the Settings menu in Noibu. Each teammate gets a permission level, either Full Access or Sessions Only, and Sessions Only users can be limited to selected domains.
Can a single Noibu account cover several storefronts?
Yes. A Noibu account can be associated with multiple domains, each with its own settings, and users switch between them from the dropdown in the top-left corner of the console. New domains are added by deploying the Noibu script and asking the Customer Success Manager to associate the domain.
Who approves changes when an agency runs the agents?
A named person at the agency reviews and merges each pull request or staged change, and the client signs off on what goes live. With Noibu, nothing ships without approval.
Will running AI agents mean a client needs its agency less?
No. Agents find problems and draft changes, but the agency still decides what fits the brand, reviews the work, merges it and stands behind the result. Agencies that run agents across clients use them to do more for each client.
Related reading
- What is an AI-native ecommerce agency?
- How to write a dev ticket from a site finding that an agency will act on
- How to connect AI agents to GitHub when you don't own the repo
See what agents find on a client store: run a free website audit →
