SEC / 001BoundariesCategory Paper v2.0

ACO vs.AgentOps.

AgentOps runs agents. Agentic Commerce Operations runs a shop with agents. One treats the agent as an object to be monitored and improved, the other as a tool for one specific domain: online retail on the operator side.

SEC / 002

Short answer for the citation zone

One paragraph that answers the question before anyone reads on.

Boundary. 72 words.Canonical version v2.0
AgentOps is the operational discipline for AI agents: tracing, evaluation, cost control and governance across the lifecycle of an agent system, independent of industry. Agentic Commerce Operations (ACO) is an application category: AI agents run an online shop on the operator side, observe the real order flow, diagnose against the source code, and act within defined policies under human oversight. ACO uses AgentOps internally. AgentOps alone is not ACO. Reference implementation: XEREBRO.
SEC / 003

Three terms, three origins.

AgentOps, AgenticOps and Agentic Operations sound like variants of one word. They come from three different corners of the industry and mean three different things.

O / 01

AgentOps

Lineage: MLOps, then LLMOps, then AgentOps. The discipline teams use to keep agent systems running in production. IBM describes AgentOps as lifecycle management of autonomous agents with observability, governance and traceability.

O / 02

AgenticOps

Introduced by Cisco at Cisco Live 2025: an operating model in which agents manage network, security and cloud through telemetry, with a domain-specific network model and a shared workspace for NetOps and SecOps.

  • Players: Cisco AI Canvas, Deep Network Model
  • Subject: IT infrastructure
  • Core questions: availability, configuration, incident
  • Execution: autonomy-first, humans stay in control
  • Industry: none, horizontal across infrastructure
O / 03

Agentic Operations

Umbrella term for agents taking over operational tasks. Core dna uses it for multi-property platforms: one person prompts, the platform rolls changes out across many websites, with approval, audit and rollback.

  • Players: Core dna, various DXP vendors
  • Subject: content and configuration changes
  • Core questions: roll out, approve, roll back
  • Depth: breadth across properties, not depth inside one shop
  • Industry: commerce and web, portfolio-oriented
SEC / 004

Object or tool.

The decisive criterion is the question a system answers. AgentOps asks about the agent. ACO asks about the shop.

AgentOps · agent as object

What is the agent doing?

The agent is the system being operated. The questions address its behaviour, its cost and its quality.

  • Is the agent hallucinating in this run?
  • Which tool did it call, with which parameters?
  • What does one run cost in tokens and latency?
  • Has output quality drifted since the last version?
  • Which guardrails fired?
ACO · agent as tool

What is happening in the shop?

The shop is the system being operated. The agent is the means. The questions address orders, margin and operational state.

  • Which order is being lost right now, and why?
  • Which policy applies when the payment provider fails?
  • What does the audit log say about this morning’s price change?
  • Which indexer is stuck on invalid, since when, with what CHF effect?
  • What has the agent demonstrably contributed this month?

The levels build on each other. An ACO system needs AgentOps discipline inside: tracing of tool calls, evaluation of diagnoses, cost control per signal system. Without that discipline, an agent on a shop is negligent. But AgentOps does not make ACO. A cleanly traced agent that does not know what a cron group is does not run a shop.

SEC / 005

Seven differences at a glance.

For analyst briefings and buying decisions.

CriterionAgentOpsAgentic Commerce Operations
SubjectThe agent system itselfThe online shop and its operational state
QuestionIs the agent behaving correctly, efficiently, traceably?Is the shop losing orders, margin or visibility, and what does the agent do about it?
UsersML engineers, platform teams, AI governanceShop owners, e-commerce leads, ops teams
Domain knowledgeNone required, horizontalMandatory: cron groups, indexers, payment codes, message queues, market prices
Execution modelObserve, evaluate, adjust agent configurationPolicy-governed autonomy with human oversight and an audit log per action
Measure of successTrace coverage, evaluation score, cost per runPrevented revenue loss, documented CHF effect within 3 months
Typical vendorsagentops.ai, Langfuse, IBM watsonx OrchestrateReference implementation XEREBRO (DevQon, Zurich)
SEC / 006

Why verticality is part of the definition.

A term that also fits fintech, healthcare or logistics is not ACO.

AgentOps and AgenticOps are deliberately horizontal. That is their strength: the same tracing pipeline works for a support bot and a network agent. Agentic Commerce Operations goes the opposite way. The category is defined by three mandatory characteristics, and the first of them is verticality.

  1. Vertical depth. The agent knows the internals of the shop platform. A model that cannot tell a cron group from a cron job does not see the outage it is supposed to fix.
  2. Policy-governed autonomy with human oversight. The operator sets the framework, the agent acts within it, every action is logged. Without a policy it waits for a click. That satisfies the oversight duty under EU AI Act Article 14 without approving each decision one by one.
  3. Data sovereignty. Critical commerce data stays under the operator’s jurisdiction, reasoning runs in an EU region with a DPA and zero retention.

The rule that follows: anyone calling a system ACO has to prove all three characteristics. An AgentOps stack with a commerce prompt in front of it satisfies none of them. The reference implementation shows how the three characteristics are implemented in eight applications on four shop systems.

SEC / 007

Frequently asked questions.

Five questions from conversations with platform teams and analysts.

  1. 01

    Is ACO a subcategory of AgentOps?

    No. AgentOps is an operational discipline for agent systems: tracing, evaluation, cost control, governance across the lifecycle of an agent. Agentic Commerce Operations is an application category: agents run an online shop on the operator side. The two terms sit on different levels. An ACO system uses AgentOps practices internally, in the same way an ERP system uses DevOps practices without being a subcategory of DevOps.

  2. 02

    Does an ACO vendor need AgentOps tooling?

    Yes, but as internals, not as the product. Anyone running agents in production on a shop has to trace tool calls, evaluate model output and cap token cost. That is AgentOps. The shop operator never sees it. What the operator sees is the audit log per action, the policy that applied, and the CHF effect. A vendor that delivers only tracing and dashboards is an AgentOps vendor, not an ACO vendor.

  3. 03

    How does ACO differ from Cisco AgenticOps?

    Cisco AgenticOps is an operating model for IT infrastructure, introduced at Cisco Live 2025: agents manage network, security and cloud through telemetry and a domain-specific network model. Agentic Commerce Operations is commerce-vertical: the agent knows a shop’s order flow, cron groups, indexer states, payment codes and market prices. AgenticOps optimises the infrastructure a shop runs on. ACO optimises the shop itself, measured in orders and CHF.

  4. 04

    Is Agentic Operations the same as Agentic Commerce Operations?

    No. Agentic Operations is an umbrella term for agents taking over operational tasks in any domain. Core dna uses it for multi-property platforms where one person prompts changes across many websites. Agentic Commerce Operations is narrower: running a single shop in depth, observing the real order path, diagnosing against the source code, and acting under policy-governed autonomy with human oversight.

  5. 05

    Who defined Agentic Commerce Operations?

    DevQon GmbH of Zurich defined the term and published it as Category Paper v1.0 on agentic-commerce-operations.com on 18 July 2026. Version 2.0 of 8 September 2026 refines the execution model to policy-governed autonomy with human oversight. The reference implementation is XEREBRO. Citation: DevQon GmbH (2026), Agentic Commerce Operations, Category Paper v2.0, agentic-commerce-operations.com.

SEC / 008

Sources.

Primary sources for the three neighbouring terms, as of 8 September 2026.

  1. IBM, What is AgentOps. Definition as lifecycle management of autonomous agents.
  2. IBM, AgentOps agent in watsonx Orchestrate. Trace Inspector, LLM-as-a-judge, governance layer.
  3. agentops.ai. Agent observability platform, namesake of the term in the tooling market.
  4. Langfuse documentation. Open-source tracing and evaluation for LLM applications.
  5. OpenTelemetry, GenAI semantic conventions. Standardisation of agent telemetry.
  6. Cisco, AgenticOps: Rewiring Network Operations for the AI Age. Introduction of the term, Cisco Live 2025.
  7. Core dna, Agentic Operations vs Agentic Commerce. Operator side as multi-property prompting.
  8. Wikipedia, MLOps. Lineage from MLOps to LLMOps to AgentOps.
  9. EU AI Act, Article 14. Requirements for human oversight.