IBM used its annual Think conference this week to unveil the next generation of watsonx Orchestrate — a platform that evolves from an AI assistant builder into a full multi-agent control plane, designed to let enterprises deploy, govern, and coordinate thousands of AI agents regardless of where they were built or which models they run on.
The timing is deliberate. Gartner predicts that 40% of enterprise applications will embed task-specific AI agents by the end of 2026 — up from less than 5% just a year ago. Yet only 2% of organisations have deployed agents at scale, largely because the tools for orchestrating and governing them simply haven't kept pace with the tools for building them.
IBM is betting that gap is where the real money sits.
What IBM actually announced
"The enterprises pulling ahead are not deploying more AI — they're redesigning how their business operates," said IBM CEO Arvind Krishna during the conference keynote. "Running AI in the enterprise requires a new operating model."
The centrepiece is watsonx Orchestrate's new agentic control plane, currently in private preview. The platform now supports agents built across different frameworks — IBM native agents, LangFlow, LangGraph, and agents using the open A2A (Agent-to-Agent) protocol — without requiring teams to rebuild what they already have.
Six core capabilities define the upgrade:
- Cross-framework agent management — bring agents from any supported framework into a single operational layer
- Observability and evaluation — tracing, runtime evaluation, and continuous performance optimisation
- Dynamic model routing — an AI Gateway routes across IBM Granite, OpenAI, Anthropic, Google Gemini, Mistral, or Llama while maintaining governance
- Enterprise security controls — centralised identity management, credential isolation, and audit logging
- Runtime guardrails — policy enforcement and compliance controls applied at execution time
- Governed agent catalogue — a marketplace of certified agents and tools with lifecycle management
Rob Thomas, IBM's SVP of Software, framed the positioning explicitly. "It's about the best agentic technology from any company in the world," he told SiliconANGLE. The platform isn't asking companies to pick IBM's models — it's asking them to pick IBM's management layer.
Why orchestration is the new battleground
The shift from "build agents" to "govern agents" isn't unique to IBM. Microsoft's Copilot has expanded into agent coordination. Salesforce rebuilt its entire platform architecture around autonomous AI agents. ServiceNow announced its own "control tower" for enterprise AI the same week as Think 2026.
But the underlying problem is real and growing. The average enterprise already runs 12 AI agents, projected to reach 20 within two years. Most were built by different teams, on different frameworks, with different governance standards — or none at all.
"Enterprises are still wrestling with fragmented data, inconsistent pipelines, and a lack of production governance," noted analyst Sid Nag of Tekonyx. The tools for building agents have raced ahead of the tools for running them reliably.
Steven Dickens of HyperFRAME Research put it more bluntly: "The market needs an agent orchestration and governance layer, ideally from a trusted third party." IBM — with its decades of enterprise credibility and no horse in the foundation model race — is positioning itself as exactly that neutral party.
The AI divide IBM is trying to exploit
IBM's framing rests on a stark reality. Despite enormous investment in AI, 56% of CEOs report seeing neither increased revenue nor reduced costs from their AI initiatives. Meanwhile, 79% of organisations face challenges adopting AI, with 54% of C-suite executives admitting the process is "tearing their company apart."
Thomas drew an analogy to early electrification. "It's useful, but it's not really redefining how the company runs," he said. "This is about moving beyond light bulbs to things that are more fundamental to how a company operates."
The implication: most businesses have scattered AI experiments producing scattered results. What they lack is the connective tissue — the orchestration, the governance, the operational rigour — that turns experiments into systems.
IBM's pitch is that watsonx Orchestrate provides that connective tissue. Whether it can deliver is another question entirely.
The sceptic's case
Analysts aren't unconvinced, but they're cautious. Patrick Moorhead of Moor Insights & Strategy acknowledges IBM's near-term opportunity but flags a limited moat — every hyperscaler is building agentic capabilities. IBM's differentiation is cross-platform neutrality, but that advantage erodes if AWS, Azure, or Google build sufficiently open orchestration layers of their own.
HyperFRAME Research raises a more practical concern: the operational retraining burden for teams managing agent drift. If agents hallucinate within deterministic workflows, entire business processes break. Orchestration without mature evaluation and correction mechanisms is just organised chaos.
And Gartner itself warns that more than 40% of agentic AI projects will be cancelled by end of 2027 due to escalating costs, unclear business value, and inadequate risk controls. The platform that manages agents still needs agents worth managing.
What this means for your business
If you're running a business with 10-50 people, you're probably not deploying thousands of AI agents yet. But the patterns IBM is responding to will reach you within 12-18 months. Here's what to watch:
The governance question arrives early. Even with three or four AI tools running across your operations — a chatbot here, an automation there, a copilot in your CRM — you'll face the same coordination problem IBM's enterprise customers face today. Who's monitoring what the agents do? Who notices when they drift? The businesses that think about this now will spend less fixing it later.
Multi-vendor is the default. IBM's explicit support for agents from any framework reflects reality: no single vendor will own your entire AI stack. The companies building agent infrastructure all assume a multi-model, multi-vendor world. Choose tools that interoperate rather than lock in.
Orchestration is a skill gap. The talent shortage isn't just in building AI — it's in managing the security and governance of agents operating autonomously in your business. This is where consultancies and managed services will earn their keep over the next two years.
IBM's announcement won't change your operations tomorrow. But it signals where enterprise technology is heading: away from building more AI, and toward making existing AI work together reliably. That's a lesson worth absorbing regardless of your scale.
Sources
- Think 2026: IBM Delivers the Blueprint for the AI Operating Model — IBM Newsroom
- Manage all your AI agents in one place with watsonx Orchestrate — IBM Blog
- IBM charts AI operating model to move enterprises beyond experimentation — SiliconANGLE
- IBM Watsonx Orchestrate and the Friction of Autonomous Agent Governance — HyperFRAME Research
- IBM Targets Enterprise AI with 'Operating Model' Push — Data Center Knowledge
- Gartner predicts 40% of enterprise apps will feature AI agents by 2026 — UC Today
