IBM's Think 2026 framework reveals a coordination problem the WEM industry has been quietly avoiding. With thousands of AI agents now operating across enterprise contact centers - each from a different vendor, built for a different task - the question of who, or what, orchestrates them has become a boardroom priority.
Contact center leaders asking how to manage AI agents across their workforce will find that most WEM platforms offer a partial answer at best. That gap has just been thrown into sharp relief by IBM.
At its Think 2026 event in Boston, IBM announced a significant expansion of its watsonx Orchestrate platform, designed to coordinate thousands of AI agents - built by different teams, running on different infrastructure - across complex business workflows. The system manages coordination, conflict resolution, and task delegation at enterprise scale. In IBM's framing, it is an operating model for the agentic enterprise.
What Is Multi-Agent Orchestration - and Why Does It Matter for WEM?
Multi-agent orchestration is the discipline of managing multiple AI agents that operate simultaneously, often with overlapping responsibilities, across shared systems and data. In a contact center context, that means coordinating virtual customer-facing agents, AI-powered QA scoring tools, automated scheduling bots, real-time coaching assistants, and forecasting models - each potentially sourced from a different vendor - as a coherent, governed system.
Most contact centers are not doing this. What they have instead is a collection of point solutions running in parallel, with limited awareness of each other and no unified control layer. Research from MIT Sloan and BCG found that while 79% of enterprises are already deploying AI in operations, 47% admit they have no strategy for managing their AI agents.
The scale of the challenge makes that figure harder to ignore. Cisco forecasts that agentic AI will handle 68% of contact center interactions by 2028. Gartner estimates the resolution rate for routine service issues will reach close to 80% by 2029. At that density, an uncoordinated AI workforce is not a productivity tool - it is a liability.
Is the WEM Industry Ready for Multi-Agent Workforce Management?
The honest answer is ‘not yet’, though several vendors are moving in the right direction.
Verint and Calabrio, now operating as a combined entity following their November 2025 merger, have published the most detailed agentic WFM roadmap in the market. Their framing shifts the WFM operator's role from queue manager to "AI technician" - monitoring model behavior, reviewing performance, setting automation guardrails.
Specific capabilities, including intraday automation that responds to demand events in seconds rather than minutes, and a Long-Term Capacity Planner built for budgeting AI headcount alongside human headcount, point to genuine architectural progress. The combined platform serves more than 10,000 organizations across 175 countries.
But Verint-Calabrio's orchestration layer, like those of its peers, is principally designed to manage agents within its own ecosystem. The harder, still-unanswered question is whether any WEM platform is equipped to govern AI agents it did not build - the NICE virtual agent running alongside the Genesys scheduling bot alongside the third-party QA tool that the enterprise bought separately.
NICE reported a 66% year-over-year increase in AI annual recurring revenue in its Q1 2026 earnings, with CEO Scott Russell pointing explicitly to growth "beyond the contact center." Genesys has announced enterprise-wide workforce orchestration capabilities that extend resource management beyond agent scheduling. Zendesk has declared the chatbot era dead and unveiled what it calls an "Autonomous Service Workforce."
Each of these announcements is directionally coherent. None of them fully solves the cross-vendor orchestration problem IBM just named.
Why Traditional WFM Models Break Under Agentic AI
The coordination deficit has real operational consequences. As CX Today has reported, classic workforce management math - the Erlang models that have underpinned contact center planning for decades - assumes random, independent arrivals. Agentic AI breaks that assumption. When an AI agent hits a confidence threshold or a policy boundary, it does not fail quietly. It escalates, often in clusters, sending bursts of complex, already-frustrated customers into human queues simultaneously.
The result, as one analysis put it, is that human agents "drop straight into escalations where the customer has already been misunderstood, bounced, or politely gaslit by a machine that sounded confident and wrong." The workforce management problem is no longer about optimizing the human queue. It is about managing the failure modes of the AI queue feeding into it.
