Blog › Agencies › Your Agency Holds the Adobe Commerce Code. How Do AI Agents Fit In?
Noibu blog

Your Agency Holds the Adobe Commerce Code. How Do AI Agents Fit In?

Your agency holds the Adobe Commerce code. How do AI agents fit in?

When your agency holds the Adobe Commerce code, AI agents fit in as a contributor to the agency's workflow, not a way around it. Noibu installs from the Magento admin with no repository access. The Bug Resolution agent finds and ranks issues, then opens a pull request in the repository the agency controls, and nothing ships until it is approved.

Last updated November 19, 2026

Plenty of Adobe Commerce merchants run the admin, the catalog and the promotions, while an agency holds the code: the custom modules, the theme, the extensions and the deploy pipeline. On a November 2026 sales call, a digital platforms manager at a retailer running Magento described the setup plainly: the in-house team has no access to its own code, and the agency's developer handles every change to the store. When AI agents come up, the first question is whether they can fit into that arrangement at all. This guide answers it for Adobe Commerce specifically, step by step.

Key takeaways

  • On Adobe Commerce, the agency's repository is usually the only route to production, and that should stay true when AI agents arrive.
  • Finding issues needs no code access: Noibu's script goes in through the Magento admin.
  • Tracing an issue into the code and opening a pull request needs repository access that the agency grants and scopes.
  • Agent pull requests should target an integration branch, get tested on staging, and merge to production only after approval.
  • Write the target branch, reviewers, off-limits code and rollback into the agency agreement before the first pull request.

Part of: What Is an AI-Native Ecommerce Agency?

What happens when your agency owns the Adobe Commerce code?

When an agency owns the Adobe Commerce code, the agency's repository becomes the only route to production. That is by design on Adobe Commerce on cloud infrastructure. Adobe's Pro develop and deploy workflow describes developing on a branch based on the integration environment, moving verified code to staging, and merging to production after testing, with each merge triggering the build and deploy.

Many agencies also connect an external repository. Adobe's GitHub integration makes the GitHub repository the code source and turns the Adobe Commerce project repository into a mirror of it, and the Bitbucket integration synchronizes a Bitbucket repository with the cloud project in the same way. Whichever host the agency uses, whoever controls that repository controls what reaches the live store.

That control is a sound arrangement, and AI agents should not route around it. The useful question for an Adobe Commerce merchant is not how to get past the agency, but which parts of the agent workflow need the code at all, and how the parts that do can flow through the agency's existing review.

What can AI agents do on Adobe Commerce without touching the code?

AI agents can find and rank issues on an Adobe Commerce store without any repository access. Noibu's help article on deploying the Noibu script on Magento adds the collection script through the Magento admin: Content, then Design, then Configuration, then the HTML Head section, where the script is pasted into the Scripts and Style Sheets field alongside anything already there. A merchant admin user can do this without opening a ticket with the agency, though the article notes that a host that blocks requests from other domains needs to allowlist noibu.com.

From that point, Noibu's agents read real shopper sessions on the storefront and checkout. They rank issues by the sessions and revenue each one touches, which is the same approach covered in how to monitor an Adobe Commerce store for errors that cost sales. The findings are available in the Noibu console and in Claude, ChatGPT and other AI tools through Noibu MCP, so a merchant can see what is breaking before deciding what to ask the agency for.

How does the Bug Resolution agent hand work to the agency's repository?

The Bug Resolution agent works inside the agency's process rather than beside it. It identifies the top fixable issues from real sessions, traces the root cause into the codebase, and creates a pull request for review. On an agency-held Adobe Commerce store, that pull request lands in the repository the agency controls, where the agency's developers see it next to their own work.

The input is the ranked issue with its evidence: the sessions affected, the error details and the revenue at stake. The output is a proposed change as a pull request, with the reasoning attached. The approval happens in two places. An agency developer reviews the diff and tests it on an integration environment and staging, and the merchant signs off on anything that changes what shoppers see or pay. Nothing ships until that approval is given, and the merge to production stays with the agency.

How an agent's change reaches an agency-held Adobe Commerce store

Purple steps are done by Noibu's agents. Every other step stays with people.

  1. 1 · Merchant

    Install from the admin

    The merchant adds Noibu's script under Content, Design, Configuration, HTML Head. No repository access needed.

  2. 2 · Agents

    Find and rank

    Agents read real shopper sessions and rank issues by the sessions and revenue they touch.

  3. 3 · Agent

    Open a pull request

    The Bug Resolution agent traces the root cause into the code and opens a pull request in the agency's repository.

  4. 4 · Agency

    Review on integration and staging

    An agency developer reviews the diff and tests it on an integration environment, then on staging.

  5. 5 · Agency + merchant

    Approve and merge

    After sign-off, the agency merges to production. The merge is what deploys.

Tracing into the code and opening a pull request is the one step that needs repository access. The agency decides how that access is scoped, for example permission to open pull requests against an integration branch without permission to merge or to push to production. The GitHub mechanics are covered in how to connect AI agents to GitHub when you don't own the repo, and keeping a steady stream of agent pull requests manageable is covered in how to keep AI-generated pull requests reviewable.

What each step needs on an agency-held Adobe Commerce store

Repository access is only needed from the pull request step onward, and the agency decides how it is scoped.
StepNeeds repo access?Who actsAdobe Commerce detail
Install the collection scriptNoMerchant admin userAdded in the Magento admin under Content, Design, Configuration, HTML Head
Find and rank issuesNoNoibu agentsRead from real shopper sessions on the storefront and checkout
Hand findings to the agencyNoMerchantA ticket with sessions, error details and revenue at stake
Trace the cause and open a pull requestYes, scoped by the agencyBug Resolution agentPull request in the repository the agency controls
Review and testYesAgency developerIntegration environment, then staging
Merge to productionYesAgency, after approvalThe merge triggers the build and deploy

What should your Adobe Commerce agency agreement say about AI agent pull requests?

The agency agreement should settle five Adobe Commerce specifics before the first agent pull request is opened. Repository ownership, intellectual property and the offboarding clause are covered in who owns the code when AI agents work on your store through an agency; the lines below are the ones that are particular to how Adobe Commerce is built and deployed.

  1. Target branch. Agent pull requests target a development branch based on the integration environment, never the production branch.
  2. Test path. Every agent change is built on an integration environment and tested on staging before anyone proposes the merge to production.
  3. Where changes live. Proposed changes go into the agency's custom modules and theme, not into core or vendor code that the agency's upgrade process depends on.
  4. Named reviewers. One named agency developer reviews the code, and one named merchant approver signs off on checkout, pricing, tracking and other shopper-facing changes.
  5. Rollback. The agency reverts the merge if a change misbehaves, and the pull request history records why each change was made.

The split between what the agency approves and what the merchant approves is worth writing down by change type. Who approves AI changes on a client's store sets out an approval matrix that drops straight into this part of the agreement.

What if the agency will not grant any repository access?

If the agency will not grant any repository access, the agents still do the finding, and the agency does the building. The merchant shares each ranked finding from the Noibu console as a ticket in the agency's own system, with the sessions, error details and revenue at stake attached, and the agency builds the change in its usual workflow. How to write a dev ticket from a site finding that an agency will act on covers what to put in that ticket.

The trade-off is speed. A findings-only handoff skips the step where the Bug Resolution agent traces the cause into the code and drafts the change, so the agency's developer does that work by hand. Many merchants start there and widen access once the agency has reviewed a few findings and trusts the evidence.

Where do the other Noibu agents fit on an agency-held store?

The Bug Resolution agent is the one that most directly meets the agency's repository, because its output is a code change. The same rule applies to every agent: work is staged for human approval, and nothing reaches the live store until a person approves it. Noibu runs six agents, Bug Resolution, Performance, CRO, ADA, A/B Testing and ROAS, and they work in the Noibu console and in Claude, ChatGPT and other AI tools through Noibu MCP.

For the agency's side of this model, where agents find and draft across every client store and people review and merge, see what is an AI-native ecommerce agency.

Frequently asked questions

Can I use AI agents on Adobe Commerce if my agency holds the code?

Yes. Noibu's collection script is added in the Magento admin, so the agents can find and rank issues without repository access. To trace an issue into the code and open a pull request, the Bug Resolution agent needs access the agency grants to its repository, and the agency still reviews and merges every change.

Does an AI agent push changes straight to my Adobe Commerce production environment?

No. The Bug Resolution agent opens a pull request for review. On Adobe Commerce on cloud infrastructure, code reaches production when someone merges it, and with Noibu nothing ships until a person approves it.

Which branch should agent pull requests target on Adobe Commerce?

Agree it with your agency in writing. A common choice is a development branch based on the integration environment, so the change is built and tested there and on staging before anyone merges it to production.

Do I have to change agencies to use AI agents on Adobe Commerce?

No. The agents feed the agency's existing review and release process. The agency keeps the repository, the code review and the merge, and the merchant keeps approval of changes that affect shoppers.

What if my agency will not give any outside access to the repository?

Use a findings-only handoff. Share the sessions, error details and revenue at stake from the Noibu console as a ticket in the agency's own system, and let the agency build the change in its usual workflow.

Summary

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.

See what the agents find on your Adobe Commerce store: run a free website audit →

Back to all blogs
‍
‍
See what Noibu’s AI agents would change on your store first.
Free website audit
Share

Keep your store running at its best, with AI agents that never clock out.