The CMDB has been declared dead more than once. Forrester argued that the CMDB should give way to an IT management graph, Flexera replied that the CMDB is not dead, and Protiviti weighed in again in July 2026. Meanwhile every major platform vendor has attached a Gen AI or agent story to the configuration record. For anyone who owns a CMDB, the useful question is narrower: what do these AI features do to configuration data that is already incomplete, and which of them have shipped?
This article reviews the vendor documentation, release notes and surveys available in October 2026. We separate what is generally available from what is beta, preview or announced, and we mark which numbers come from vendors. Where we found no evidence, we say so. If you want the hardware side of this story, read our companion article on AI in hardware asset management tools.
What AI can and cannot do for a CMDB
AI helps with the work around the record: finding duplicates, proposing relationships, explaining what depends on what, drafting a change risk assessment and answering questions in plain language. It does not make the record true. If discovery misses a cloud account, or ownership fields are blank, an assistant will describe the incomplete picture with the same confidence as a complete one.
The survey evidence points the same way. An HBR and Cloudera study published on 5 March 2026 found that only 7 percent of respondents say their data is fully ready for AI. The ITSM.tools State of Agentic AI in ITSM 2026 survey, sponsored by HCLSoftware, put poor data quality first among barriers at 32 percent, ahead of governance and compliance at 30 percent. Gartner said in February 2025 that organizations will abandon 60 percent of AI projects that lack AI-ready data through 2026. These are not CMDB figures, but the CMDB is the data AI agents are told to use for IT questions.
The tools, one by one
ServiceNow
ServiceNow’s Australia release, generally available from 5 May 2026, is the most concrete set of CMDB AI features we found. It includes AI-assisted deduplication, AI Service Mapping, Service Mapping tools exposed through MCP, intelligent search in the CMDB Workspace and CSDM 5.0. ServiceNow also announced a Context Engine and Autonomous Data Governance for the second half of 2026. We did not find evidence that either was generally available at the time of writing, so treat them as roadmap.
BMC Helix
BMC Helix CMDB 26.2 adds a CMDB Governor, a Data Quality Agent and a Recommendation Agent. Agent-to-agent communication is in controlled availability, which in practice means you ask BMC for access. The design is notable because it frames the agents as data stewards, not as chat interfaces.
Freshservice and Device42
Freshworks completed the integration of Device42 discovery into Freshservice, available since 31 March 2026. Discovery and dependency mapping are the foundation rather than an AI feature in themselves, but they are what any AI layer on top will reason about. We found MCP access announced for Freshservice but no AI-specific CMDB agent.
Atlassian Jira Service Management and Assets
Atlassian’s Teamwork Graph MCP server is in open beta, and AI Change Risk Assessment became generally available in Jira Service Management on 7 October 2026. Atlassian says its graph-based approach produced 44 percent more accurate answers with 48 percent fewer tokens. That is a vendor-reported figure from vendor testing and we could not find the method. Atlassian also cites Forrester research that only 18 percent of organizations have their knowledge connected to a context graph.
Observability platforms
Dynatrace made its MCP Server generally available in March 2026, Microsoft has an Azure Resource Manager MCP server in public preview, and Datadog offers an MCP server and its Bits AI SRE agent. These do not replace a CMDB, but they are increasingly the source of truth for runtime topology, and an agent that can query both a CMDB and an observability platform will notice where they disagree.
Smaller and open source options
Virima and Faddom both appear in analyst and vendor material on AI for configuration data, and Faddom was named a sample vendor in Gartner’s 2026 Hype Cycle for AI in ITSM. For open source, iTop has a community MCP server at version 0.0.1. That is early, and we would call it an experiment.
Two things stand out. Deduplication, mapping and quality agents are where shipped features cluster, which fits the problem: those are the jobs people already do badly by hand. And MCP is the only row with nearly every tool in it. That is a change in kind. A CMDB with an MCP server is a data source for agents you did not build.
Use cases in detail
Duplicate and stale record cleanup
Every CMDB owner has a duplicate problem. The same server appears once from a cloud connector, once from agent discovery and once from a manual import, each with a slightly different name. Rule-based reconciliation handles the cases you thought of. AI-assisted matching can propose merges for the cases you did not, with a confidence score for a person to accept. That is a good fit, because a wrong merge is visible and reversible if the process requires review.
Mapping services and dependencies
AI Service Mapping proposes which components make up a business service. The benefit is speed on the first draft. The risk is a map that looks complete and is not, which is worse than an empty map because nobody questions it. Validate against traffic or observability data, which is exactly where the MCP connections to platforms such as Dynatrace and Datadog help.
Data quality agents
BMC’s agents and ServiceNow’s planned autonomous governance aim at the same target: find fields that are blank, stale or inconsistent, and propose or make a fix. Start with proposals only. An agent that writes to production configuration records needs a defined scope, an audit trail and a rollback path.
Change risk
Change approval is where a CMDB earns its keep and where a bad CMDB does the most damage. Atlassian’s AI Change Risk Assessment and similar features score a change using relationships and history. The score is only as good as the dependency data. If you want to see how this connects to processes, read our article on aligning ITSM processes with tools.
Questions in plain language
Asking which services depend on a given database, or which servers run an unsupported operating system, is the lowest risk and most visible use case. It is also where the quality of your data shows immediately. Expect users to ask a question they know the answer to, and to stop trusting the tool when it differs.
Agents that use the CMDB
The most consequential use case is not inside the CMDB at all. An incident agent, a change agent or a security agent consults the configuration record to decide what to do. A practitioner post on the ServiceNow community on 4 June 2026 suggested sorting agent actions into three groups: read-only, needs validation, and needs human approval. It is an opinion, not a standard, but it is a sensible way to start.
The benefits, and what is actually evidenced
We found no independent study measuring the benefit of AI features in a CMDB. What exists is vendor claims, such as the Atlassian accuracy and token figures, and survey data about the foundation. A Gartner figure that Flexera cited in an October 2026 release puts CMDB success at 29 percent. We could not confirm it at the original source, so we are citing it as secondhand. If it is even roughly right, the opportunity for AI to help is large and so is the risk that it will not.
The realistic benefits are less glamorous than the marketing: faster cleanup of duplicate and stale records, a quicker first draft of service maps, fewer blank fields, quicker impact analysis during incidents and changes, and a way for people outside the CMDB team to ask questions without learning a query language. Measure each against a baseline you take before the pilot.
How it is evolving
From records to graphs. The argument in Forrester’s work is that the flat configuration item table is the wrong model for AI, and that a graph of IT entities and relationships fits better. Atlassian’s Teamwork Graph and ServiceNow’s announced Context Engine are both attempts at this. Flexera’s counterargument is that the CMDB still matters because it is the governed system of record. Both can be true: the graph may be how agents read the data, while the CMDB remains where ownership and lifecycle are managed.
From rules to stewards. BMC’s Governor and ServiceNow’s planned autonomous governance move data quality from a quarterly cleanup to a continuous process. The question each buyer should ask is who is accountable when an agent’s change is wrong.
From dashboards to protocols. MCP and agent-to-agent communication make configuration data available to agents from other vendors. That is useful and it enlarges your access control problem. Treat an MCP server on your CMDB like an API you expose to the whole company.
The analyst view is cautious. Gartner said in June 2025 that more than 40 percent of agentic AI projects will be canceled by the end of 2027, and in a 29 September 2026 release that 70 percent of organizations will abandon agentic AI built by vendor forward-deployed engineers by 2028. Its May 2026 autonomy framework expects many organizations to demote autonomous agents to lower levels by 2027.
Risks and prerequisites
Ownership. Every CI class needs a named owner. Without one, an agent’s proposed fix goes to nobody.
Discovery coverage. Cloud accounts, containers and SaaS are the usual gaps. A model trained on, or an agent reading, a partial inventory has no way to flag what is missing.
Data model alignment. CSDM 5.0 and similar models help only if you have adopted them. Ask how the AI feature behaves with a customized schema.
Write access. Give agents read access first. Add write access for narrow classes of record, with logging and rollback.
Availability and cost. Several features above are beta, controlled availability or announced. Confirm which edition and licence includes each one, and what consumption they bill against.
A readiness checklist before you pilot
- Is there a named owner for each major CI class and each business service?
- Do you know your discovery coverage for cloud, on-premises and SaaS?
- Have you measured duplicate, stale and incomplete record rates?
- Can you list which agent actions are read-only, which need validation and which need approval?
- Is access to any MCP server limited, authenticated and logged?
- Do you have a baseline for incident impact analysis time and change failure rate?
Where a maturity assessment fits
A CMDB maturity assessment tells you which of those answers are yes before you pay for the feature. Our Asset and Configuration Management maturity assessment covers ownership, data quality, discovery, governance and integration. If your concern is wider than configuration data, the ITSM and ITOM maturity assessment covers the surrounding processes, and you can choose another category. For more background, see how a HAM and CMDB maturity assessment helps an organization, our ITOM assessment article and the guide to HAM and CMDB tools.
Frequently asked questions
Is the CMDB still relevant if AI agents can query tools directly?
Yes, as the governed record of ownership, relationships and lifecycle. Agents can read from many sources, but someone still has to decide which one is authoritative.
Which CMDB has the most shipped AI?
ServiceNow has the most generally available CMDB AI features in our review, followed by BMC. Several of the announced capabilities are not yet available.
Should an AI agent be allowed to change CMDB records?
Not at the start. Begin with read-only and proposal modes, add validation by data owners, and keep human approval for changes that affect production records or trigger a change.
Do we need to adopt a graph database?
Not necessarily. Several vendors are building graph layers on top of existing records. What matters is that relationships are accurate and owned.
Find out how ready your configuration data is
Take the Desqcon online maturity assessment for asset and configuration management, or talk to a senior consultant about a CMDB health check before you add AI.
Take the CMDB Maturity AssessmentITSM and ITOM AssessmentTalk to a Senior Consultant
Sources and further reading
- ServiceNow community, Australia release notes for CMDB and Service Mapping, May 2026, and the practitioner post on agent action classes, 4 June 2026
- BMC Helix CMDB 26.2 release documentation
- Freshworks, Freshservice and Device42 availability announcement, 31 March 2026
- Atlassian, Teamwork Graph MCP server beta and Team ’26 Europe announcements, October 2026
- Dynatrace MCP Server documentation, March 2026; Microsoft Learn, Azure Resource Manager MCP server preview; Datadog MCP Server and Bits AI SRE documentation
- iTop MCP server on GitHub, version 0.0.1
- Forrester, The CMDB is dead, long live the IT management graph; Flexera, CMDB isn’t dead; Protiviti, 14 July 2026
- HBR Analytic Services and Cloudera, 5 March 2026
- ITSM.tools, State of Agentic AI in ITSM 2026
- Gartner press releases: 26 February 2025, 25 June 2025, 26 May 2026, 29 September 2026
