That’s it: IBM now also has its own “end-to-end control plane” for AI agents.
The company presented it that way at TechXchange 2026. Under the brand Agent Identity, it offers an actionable registry and an authorization engine that work with HashiCorp Vault for secret management and with watsonx Orchestrate for agent discovery (feature in preview, similar to sandboxing).
SaaS-Only to Start
While waiting for the self-hosted version and the IVIA add-on slated for December, Agent Identity is available within IBM Verify’s SaaS IAM.
Billing is tiered, based on monthly active users thresholds. It also depends on the volume of tokens issued.
Each user—every agentic entity or anyone who interacts with an agent—has a soft limit of 100 tokens per month, after which excess billing applies.
Consumption is, however, aggregated across all users over a year-long period or the contract duration. Each tier has a conversion ratio into RU (resource units).
IBM intends to provide “soon” an add-on to accelerate transaction throughput.
From Build to Run, a Bit More AI for the Mainframes
In the realm of agents, there is also involvement in the mainframe universe, with IBM Z Atlas.
Big Blue dedicates a long statement of intent to this offering. The simple promise: to map z/OS environments to foster AIOps. In other words, to provide the context needed by AI agents that will assist with change planning and modernization.
With IBM Z Atlas, we are in the build phase. For the run, there is, among other things, watsonX Assistant for Z. The next version (4.0), announced for November 6, is re-architected around the Agent Skills standard. As a result, embedded sub-agents become consumable as skills by the main agent. In parallel, the Knowledge Transformer component (preparation of unstructured data for RAG) moves to GA.
Another statement of intent noted shortly before TechXchange: IBM promises, without giving a timeline, an interface unifying the consoles of several logical partitions.
