Managed Agents vs Managed Services: What Changes for Commerce Teams?
Managed services sell staff time: an agency team works through a ticket queue on a monthly retainer. Managed agents sell a running loop: AI agents find problems and draft changes, and named people at the agency review, merge and measure them. For commerce teams, the job shifts from writing tickets to approving finished work.
TL;DR
- Managed services bill for people and hours. Managed agents bill for a loop that ships reviewed changes.
- The agency does not go away. It becomes the reviewer and merger of record.
- Commerce teams stop writing tickets and start approving drafted changes with evidence attached.
- Ask any partner where the agents get their data, how changes reach review, and how impact is measured.
The phrase managed agents vs managed services comes from a shift in how commerce agencies sell their work. This post is part of Noibu's series on what an AI-native ecommerce agency is, and it focuses on one question engineering and ecommerce leaders now ask partners: what actually changes when an agency moves from a services retainer to a model built on AI agents?
What is the difference between managed agents and managed services?
Managed services is the retainer model most commerce teams know. A merchant pays an agency a monthly fee for a team of developers, and that team works through a queue of tickets the merchant writes. The merchant pays for capacity whether or not the queue produces measurable results.
Managed agents is a model in which AI agents do the repeatable parts of that work across every client the agency supports. The agents watch the store, find problems, and draft changes. The agency's engineers review those drafts, decide what merges, and own the outcome. The merchant pays for a loop that keeps producing reviewed improvements.
Managed services vs managed agents, side by side
Managed services
Unit sold
Hours or seats on a monthly retainer
Who finds work
The merchant, through tickets and reports
Who builds
Agency developers, one ticket at a time
Who approves
The agency, often after the work is done
What you see
Hours used and tickets closed
Learning
Stays with the people on the account
Managed agents
Unit sold
A running loop of reviewed changes
Who finds work
Agents reading real shopper data
Who builds
Agents draft; engineers refine
Who approves
A named person reviews every change before merge
What you see
Changes shipped and measured lift
Learning
Codified in shared skills and runbooks
John Duncan, CEO of the Salesforce Commerce Cloud agency 64labs, set out the argument in Managed Agents, Not Managed Services (August 31, 2026). He describes enterprise managed services as mostly staff augmentation and writes that a named human reviews every merge in the agent model he runs.
Why are agencies moving from managed services to managed agents?
Agencies are moving because the work inside a services retainer is increasingly the kind AI agents can draft. Triage, reproducing problems, small front-end changes, and performance tuning used to take developer hours. When agents draft that work, an hours-based retainer stops matching the value delivered.
Duncan's article makes three arguments. First, skills and runbooks live in shared infrastructure, so improvements spread across clients: in his words, “A lesson learned on one client's build ships to every client's build.” Second, he estimates the cost reduction at 50 to 80 percent rather than 20 percent; that range is his own estimate, not an audited benchmark. Third, he expects lower unit costs to raise the amount of work clients commission, such as more experimentation and personalization.
For merchants, the third point matters most. A managed agents model does not have to mean a smaller agency relationship. It can mean the same budget buys more shipped, measured changes.
What changes for commerce teams under managed agents?
Five things change for the ecommerce and engineering teams on the merchant side. Together they are the practical checklist for judging whether an agency's managed agents offer is real.
- Where work comes from. Work starts from evidence in real shopper sessions, not from someone noticing a problem and writing a ticket. Over 90% of website errors are never reported, so a ticket queue misses most of what breaks.
- What arrives for review. Teams receive drafted changes, such as pull requests with the cause and evidence attached, instead of estimates for future work.
- Who approves. Approval moves before the merge. A named reviewer at the agency or in-house signs off on every change.
- How progress is reported. Reports show changes shipped and their effect on orders and revenue, not hours consumed.
- Where knowledge lives. What the agency learns is written into skills and runbooks rather than held by the people on the account.
Where people sit in a managed agents loop
1
Find
Agents read errors, speed and conversion data from real shopper sessions.
2
Build
Agents draft the change as a pull request or a staged edit.
3
Review
The agency or in-house engineer reviews, edits or rejects it.
4
Ship
The approved change merges through the normal release path.
5
Measure
Agents report the result in orders and revenue.
Review volume becomes the new constraint. When agents draft more changes than a team can read, review queues back up. Noibu's guide to keeping AI-generated pull requests reviewable covers batching, ranking and approval limits.
Do managed agents replace the agency?
Managed agents do not replace the agency. The agency stays the reviewer and merger of record, because someone accountable has to decide what reaches a live store. What changes is how the agency's people spend their time: less on finding and drafting, more on review, architecture and the work that needs judgment.
Agencies working this way report doing more for each client, not less. In Noibu's field work, one agency wired every client deploy into Noibu and had run more than 100 Noibu MCP sessions by mid-October 2026, using the agents to feed its own release process. Blue Badger, a Noibu agency partner, increased client retainers by 20% with Noibu.
When the agency controls the repository, the setup question is practical rather than political. Noibu's guide on how to connect AI agents to GitHub when you don’t own the repo walks through access, review rules and who merges.
What should engineering leaders ask before moving to managed agents?
Engineering leaders should test a managed agents offer the same way they would test any change to the deploy path. These questions separate an agent loop from a services retainer with AI added on top.
- Where do the agents get their data? Agents that read real shopper sessions, errors and Core Web Vitals find problems a staging environment never shows.
- How do changes reach review? Look for pull requests in the repository you already use, with evidence attached, and nothing merged without approval.
- Who is the named reviewer? Every merge should have one accountable person, at the agency or in-house.
- How is impact measured? Each shipped change should be measured in orders, conversion or revenue after release.
- What happens to what the agency learns? Duncan suggests asking whether a partner is codifying its knowledge or keeping it in people's heads.
Measurement is where many retainers fall short. Noibu's guide to checking whether a website change actually improved conversion shows how to compare before and after a release. For governance and cost controls, see the AI agent governance and token-cost checklist.
Where does Noibu fit in a managed agents model?
Noibu supplies the agents in a managed agents model, while the agency or in-house team stays in charge of review and release. The Bug Resolution agent finds the errors shoppers hit that cost the most, traces the root cause in the codebase, and opens a pull request for review. The Performance agent measures Core Web Vitals from real shopper traffic, ranks the slowest page groups, identifies causes and stages changes.
Noibu's CRO, ADA, A/B Testing and ROAS agents run the same loop on conversion, accessibility, experiments and paid traffic. Every agent follows find, build, approve, ship and measure, and nothing ships without approval. Agencies and merchants can work with the agents in the Noibu console or through Noibu MCP in AI tools teams already use, such as Claude, ChatGPT, Microsoft Copilot, GitHub Copilot and OpenAI Codex.
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
What is the difference between managed agents and managed services?
Managed services sell agency staff time on a monthly retainer, usually spent working a ticket queue. Managed agents sell a running loop in which AI agents find problems and draft changes, and named people at the agency review, merge and measure the results.
Do managed agents replace my ecommerce agency?
No. In a managed agents model the agency stays the reviewer and merger of record. Agents take on the finding and drafting, and the agency spends its time on review, judgment and the harder engineering work.
Who approves changes in a managed agents model?
A named person approves every change before it merges. With Noibu, nothing ships without approval, and agent changes arrive as pull requests or staged edits that the agency or in-house team can review, edit or reject.
Will managed agents cost less than a managed services retainer?
Some agency leaders expect large savings, but the figures published so far are estimates rather than audited results. A more useful test is how many reviewed changes reach the live store each month and what each one did to orders and revenue.
How do I start with managed agents if my agency owns the repo?
Bring the agency in before the first pull request lands, connect the agents to the agency-owned repository with review rules in place, and agree who merges. A free website audit shows what the agents would find on the store first.
Related reading
- What is an AI-native ecommerce agency?
- How to connect AI agents to GitHub when you don’t own the repo
- How to keep AI-generated pull requests reviewable
See what agents would find on your store: run a free website audit →
