Kyndryl is betting that the next big enterprise headache isn't adopting AI, it's controlling the autonomous agents already running wild inside corporate networks.
Infrastructure services provider Kyndryl has launched a new practice specifically designed to help businesses manage AI agents, a clear signal that the enterprise AI conversation is shifting from excitement about deployment to the harder question of governance. CEO Martin Schroeter, speaking with Bloomberg Technology's Tim Stenovec, framed the offering as essential for companies looking to extract real returns from their AI investments rather than watching those tools create new operational risks.
The timing is not accidental. Enterprise adoption of AI agents has accelerated sharply over the past year, with companies deploying autonomous software that can make decisions, trigger workflows, and interact with external systems without direct human oversight. The promise is efficiency at scale. The reality, for many organizations, is a growing inventory of agents that no single team fully understands or controls.
Kyndryl's move addresses a gap that has been widening since generative AI went mainstream. Companies spent 2023 and much of 2024 racing to build and deploy AI tools. Consulting firms like McKinsey, Accenture, and Deloitte rode that wave, helping clients pilot hundreds of use cases. But the second-order problem is now arriving: once you have dozens or hundreds of AI agents operating across finance, supply chain, customer service, and HR, who makes sure they are not contradicting each other, accessing data they should not touch, or making decisions that violate company policy?
This is where Kyndryl sees an opening. The company, which was spun off from IBM's managed infrastructure services business in 2021, has been working to reposition itself around higher-value services. Its core business involves managing the complex IT environments that large enterprises depend on, everything from mainframe operations to cloud migrations. Adding AI agent management to that portfolio is a natural extension of the kind of work Kyndryl already does, but it also positions the firm to capture spending that might otherwise go to cloud consultancies or hyperscaler partners like AWS, Microsoft, and Google.
The market opportunity is substantial. Global spending on AI systems is projected to surpass $300 billion by 2026, according to IDC forecasts, and a growing share of that budget is shifting from model development toward operational management and governance. Gartner has separately estimated that by 2028, one-third of interactions with generative AI services will involve AI agents performing multi-step tasks autonomously. That prediction carries an implicit warning: organizations without robust oversight frameworks will face compliance failures, data leaks, and reputational damage.
Schroeter's emphasis on control reflects a broader anxiety among CIOs and CTOs. The initial thrill of deploying large language models has given way to practical concerns about cost management, model drift, data provenance, and the unpredictable behavior of autonomous systems. High-profile incidents, including cases where AI chatbots fabricated legal citations or provided incorrect medical information, have reinforced the message that unmanaged AI is a liability, not an asset.
What This Means for the Market
For Kyndryl, success in this area would validate its post-spinoff strategy of moving beyond legacy infrastructure into higher-margin advisory and managed services. The company reported annual revenues of around $15 billion in its most recent fiscal year, but has faced pressure to demonstrate growth in a competitive market. AI governance could become a meaningful differentiator if enterprises decide they need dedicated partners for this function rather than relying on hyperscaler tools alone.
The competitive landscape is still forming. Microsoft has been building agent management capabilities into its Copilot platform. Salesforce recently launched Agentforce, its own framework for deploying and controlling autonomous AI agents across business functions. Startups like CrewAI and LangChain are also angling for pieces of this market, offering developer tools designed to make agent orchestration more transparent and manageable.
Kyndryl's advantage, if it has one, is its deep relationships with the large enterprises that run the world's most complex IT environments. These are companies that often have decades of accumulated technology debt, hybrid cloud architectures, and regulatory obligations that make straightforward AI adoption impossible. They need partners who understand their existing systems, not just the latest models.
The real test will be execution. Governance and management services are only as valuable as the outcomes they produce. If Kyndryl can demonstrate that its oversight reduces AI-related incidents, lowers operational costs, and accelerates the time to measurable ROI, the offering will resonate. If it becomes another layer of consulting abstraction, enterprises will quickly look elsewhere.
Watch for two signals in the coming months: whether Kyndryl announces specific enterprise customers for this service, and whether competitors respond with similar managed offerings of their own. The enterprise AI market is entering its most consequential phase, the phase where deployment gives way to discipline. Companies that master control will be the ones that actually profit from the technology they have spent billions acquiring.
