How IBM's June 2026 platform update brings centralized governance, live scheduling, a reusable agent catalogue and tighter data protection to enterprise AI operations
IBM watsonx Orchestrate's June 2026 mid-month release marks a shift in emphasis. Earlier releases were largely about giving builders more ways to construct agents — document extractors, flow controls, prebuilt connectors. This release is about what happens after the agents are built: how they are governed, secured, monitored, scheduled, shared and protected once they are running in production.
The headline additions are a new Agentic Control Plane dashboard, a Security Control Center, an overhauled analytics experience, natural-language task scheduling in chat and the ability to publish agents to a reusable tenant catalogue. Alongside these, IBM has shipped a steady stream of smaller but consequential updates: parallel execution in flow builder, sensitive-data masking, simplified credentials management and an expanded roster of partner agents and MCP servers. Together, these changes suggest a platform maturing from an agent-building tool into something closer to an operating system for enterprise AI.
The most consequential addition is the Agentic Control Plane, a centralized operational dashboard that gives administrators system-wide visibility into agent health, performance and governance. It is generally available in some environments and in private preview in others, with availability varying by cloud platform and region.
The dashboard is organized around four capabilities that map directly to how operations teams work:
Needs Attention — prioritized alerts for operational issues (credential problems, configuration blockers), incidents (production failures, negative feedback) and insights (evaluation gaps, quality issues)
Jump Back In — quick access to recent activity across all tenant users, including agent edits and evaluation runs
Agent Analytics — live metrics on message volume, failure rates, latency and unique users
Embedded Operations Agent — a natural-language interface for investigating performance issues and troubleshooting, without needing to manually dig through logs
For organizations running more than a handful of agents, this is the difference between reactive firefighting and proactive fleet management. The embedded operations agent in particular signals where IBM sees this heading: using AI to operate AI, rather than requiring a human to interpret every trace.
Available in AWS deployments, the new Security Control Centre addresses a problem every enterprise deploying agents eventually runs into knowing exactly which agents exist, what they can access and who authorized it. It consolidates authentication and authorization visibility into one place, organized across three views.
Agents tab — every native and imported agent, with its accessible connections, tools, child agents and OAuth permissions, searchable by ID, name or creator email
Connections tab — configuration of authentication methods, credential models and draft versus live environment values for each integration
Team credentials tab — management of shared credentials used across multiple users, by authentication type
The stated purpose is direct: identify unauthorized or shadow agents, monitor credential health and verify that access matches intent. As agent sprawl becomes a real governance risk, this kind of centralized audit surface stops being a nice-to-have and starts being a prerequisite for production deployment.
Sitting alongside the dashboards, the control plane also introduces governance controls that operate at the infrastructure level, also limited to AWS deployments for now. Three capabilities stand out:
Policy management — pre-built policies for output limits, secrets detection, rate limiting and content filtering, enforced automatically at runtime against agents, tools and models
Credential health monitoring — automatic alerts when connections have missing or invalid credentials, before they cause tool failures in production
Content guardrails — six categories of runtime content filtering (hate/abuse/profanity, harm, violence, jailbreak attempts, social bias and sexual content), powered by IBM's Granite Guardian models
Overview analytics — conversations, unique users, average duration and feedback trends across all agents, filterable by 7, 15 or 30 days
Single-agent dashboard — token counts, message latency, usage trends and LLM-as-a-judge scores for hallucination, helpfulness and toxicity
Conversation history — individual conversations with full metadata, feedback and evaluation scores, expandable to complete message context
Execution trace debugging — step-by-step visualization of input parsing, reasoning, tool invocation and response generation, with latency at each step
This is governance moving from a documentation exercise to an enforced, automated layer. Policies that used to live in a compliance wiki now run inline, at the moment an agent or tool executes.
The new Analytics experience, also AWS-only for now, replaces high-level dashboards with a progressive investigation flow: start broad, narrow down and end at the exact step in an agent's reasoning that caused a problem.
Notably, IBM is deprecating the legacy Analytics experience for AWS deployments on July 10, 2026 — just days from now. Teams on AWS should treat migration to the new experience as immediate, not optional.
Task scheduling moves from something builders configure in a UI to something anyone can request in natural language, directly in chat. Once scheduling is enabled for a workflow in flow builder or an agent in agent builder, the necessary scheduling tools become available automatically.
Create schedules conversationally, with an interactive form that walks through recurrence (daily, weekly, monthly, hourly or custom) and start/end dates
Manage existing schedules — view, edit or delete them — through the same conversational interface
Receive results directly in the chat thread after each run, with the agent inferring missing inputs from context where possible
This is a small feature with outsized practical value: timesheet reminders, monthly reports and system-monitoring checks no longer require a separate automation tool. They can be set up the same way a user would ask an agent to do anything else.
Two related features address a problem that emerges once an organization has more than a few builders: how do you stop everyone from rebuilding the same agent? Publishing to catalog lets administrators share stable, versioned agents — complete with name, description, up to three categories and a custom icon — for reuse across the tenant. Semantic versioning keeps development isolated from what is in production and change logs document what moved between versions.
The companion feature, artifact hierarchy visualization, gives builders an interactive map of an agent's composition before they adopt it — its collaborator agents, dependencies and tools, viewable in vertical or horizontal layouts. Currently limited to prebuilt agents, it answers a question builder previously had to reverse-engineer: what exactly does this agent depend on?
Together, these features push Watson Orchestrate toward an internal marketplace model, where proven agents compound in value rather than being reinvented per team.
Two updates target builders working directly in flow builder. For each controls can now run in parallel rather than strictly sequentially, which matters for workloads like document batches, multi-reviewer approvals or concurrent API calls where items don't depend on one another. The performance gain scales with list size and parallel user activities extend the same idea to workflows involving multiple human reviewers.
Separately, pro-code workflows built with the ADK can now be opened directly in flow builder as read-only visualizations — useful for inspecting structure and runtime behavior with the flow inspector — or converted into fully editable low-code flows for teams that want to continue development in the visual builder. This narrows the gap between the two authoring paths that IBM has maintained since the platform's early releases.
Data protection gets three separate but related upgrades this release. Flow builder now supports masking of sensitive fields — inputs, variables, node outputs, form fields — with options for full masking, partial masking (first or last four characters visible) or custom regex patterns, plus a mask-while-typing mode. Masked values stay hidden in chat history, the flow inspector and observability traces, which matters for anything touching passwords, API keys or payment details.
Credentials management has also been reorganized: team credentials (shared across users) now live in the Security Control Centre, while member credentials (individual user identity) move to personal profile settings — a cleaner separation than the previous single location. And SSO now supports multiple OAuth methods — including On Behalf Flow and token exchange — for Zendesk, SharePoint, Outlook and Teams, giving administrators more flexibility to match authentication to different user groups or security contexts.
Text extraction now supports specifying a page range rather than processing an entire document. For large documents where only a handful of pages are relevant — the first few pages of a contract, or a specific section of a report — this avoids the cost and noise of extracting content nobody needs. It's a narrow feature, but one that speaks to document workflows moving from proof-of-concept to production, where processing efficiency starts to matter at scale.
Embedded chat interfaces can now be customized with injected HTML at six defined locations — below the header, above the input bar, above the welcome message, below starter prompts, inside the AI disclaimer tooltip and in the centre of the chat area — via a new getWriteableElement() method. This opens the door to banners, announcements, image carousels and interactive buttons without waiting on IBM to add native support for each use case.
The partner catalogue continues to grow, with new native agents (Questionnaire Generator, JD Generator), a new external agent (GovPulse Intelligence Agent) and new MCP servers from PagerDuty, Visier and Workable. On the API side, agent conversations can now use configurable context compaction to prevent overflow in long-running sessions and all proxy error responses follow a standardized JSON structure with stable error codes and transaction IDs — a small change that meaningfully improves troubleshooting for anyone building against the API directly.
Several threads run through this release:
Operations is the new frontier. The Agentic Control Plane, Security Control Centre and rebuilt Analytics experience all point the same direction: once organizations have agents in production, the hard problems shift from building to operating, securing and troubleshooting them at scale.
Governance is centralizing, not distributing. Policies, guardrails and credential health checks now run as enforced infrastructure rather than living in separate compliance processes. This reduces the gap between what an organization intends and what actually happens at runtime.
Reuse compounds value. Catalogue publishing and artifact hierarchy visualization both assume that the biggest efficiency gains come not from building more agents, but from making existing ones discoverable, trustworthy and easy to adopt across teams.
Data protection is table stakes, not a differentiator. Masking, reorganized credentials and multi-auth support all address the unglamorous but essential work of making sure sensitive data doesn't leak through chat logs, traces or shared connections.
For CIOs and automation leaders, the pattern is consistent with where mature enterprise software categories tend to go: the building tools stabilize and the centre of gravity moves to operating what's been built responsibly and at scale.
Organizations running or evaluating Watson Orchestrate should weigh several practical steps:
Migrate off legacy analytics now. If you're on AWS, the old Analytics experience is deprecated July 10, 2026. This is not a distant deadline — confirm your dashboards and alerts are rebuilt on the new experience before then.
Stand up the Security Control Centre early. Use it to audit existing agents for shadow deployments and unauthorized connections before adding new ones. It's easier to establish a clean baseline now than to retrofit governance onto an unaudited fleet.
Pilot scheduling on genuinely repetitive work. Timesheet reminders, recurring reports and monitoring checks are low-risk places to validate the new chat-based scheduling before relying on it for anything business-critical.
Publish your most stable agents to the catalogue. Identify the two or three agents most duplicated across teams and formalize them as versioned, catalogue-published assets first.
Apply masking to anything touching credentials or PII. Audit existing flows for fields that should be masked — don't wait for an incident to surface the gap.
Track the MCSP 2.0 migration if you're on AWS. Selected regions migrate from MCSP 1.0 to 2.0 in July 2026 during scheduled maintenance windows. No action is required, but plan for the notification and confirm downstream integrations that depend on authentication behaviour.
Want to explore how these Watson Orchestrate capabilities could advance your organization's automation strategy? Let's discuss your specific requirements and deployment context.