Research Finder
Find by Keyword
Can Kyndryl’s AI Agents Really Replace Mainframe Pros?
Kyndryl bets big on autonomous agents to solve the mainframe talent crisis, but handing the keys to "agentic" code brings a massive trust paradox that CIOs might not be ready for.
25/11/2025
Key Highlights:
Kyndryl has launched a new "Agentic AI Framework" specifically targeting IBM z/OS environments to automate complex, multi-step mainframe tasks.
The move is a direct response to the severe skills shortage, with the company’s own research claiming 70% of organizations lack the multi-skilled talent needed for modernization.
New services integrate deeply with Kyndryl Bridge and IBM watsonx Assistant for Z to create a "digital workforce" capable of troubleshooting and optimizing legacy code.
While the promise of $12.7 billion in cost savings is enticing, the security and governance risks of autonomous agents on mission-critical systems remain the elephant in the room.
The News
Kyndryl has officially rolled out a suite of "agentic AI" services and a dedicated framework designed to bring autonomous automation to the mainframe. The announcement details how these new AI agents can plan, execute, and verify complex tasks on IBM z/OS systems, aiming to reduce the operational burden on shrinking human teams. You can find out more by clicking here to read the press release.
Analyst Take
I must admit, I am rather taken by the audacity of this move, particularly given the sheer weight of the data behind it. With over 6 million MIPS under management—a footprint safely four times larger than its nearest IT infrastructure competitor—Kyndryl holds an operational vantage point that is mathematically impossible for others to replicate. For years, the industry has been plastering "AI" onto mainframe modernization in a way that felt like putting a spoiler on a station wagon—purely cosmetic. But this, at least on first blush, looks substantively different; Kyndryl are leveraging that massive installed base to architect a system where the AI moves from "assisting" to "doing," a pivot that is only viable when you have the deep operational history to train the models correctly.
What was Announced
Specifically, Kyndryl introduced an Agentic AI Framework and a set of managed services tailored for the mainframe. The offering is designed to integrate with Kyndryl Bridge, its open integration platform, and leverages IBM watsonx Assistant for Z.
Autonomous Orchestration: The framework is architected to allow AI agents to "break down" high-level intent (e.g., "optimize this batch process") into specific technical steps, execute them, and learn from the results.
Skills Augmentation: A core component is the Kyndryl AI Assistant for Z, which aims to deliver a "knowledge base" of mainframe expertise to less experienced developers, effectively democratizing the specialized knowledge usually locked in the heads of retiring baby boomers.
Operational Integration: The services are designed to sit within the hybrid IT control plane, meaning these agents can theoretically see and act across both the modern cloud stack and the legacy z/OS backend.
Analyst Take
Here is where I get a bit skeptical, though. The proposition is brilliant on paper: swap your retiring experts for autonomous agents that don't sleep and don't retire. However, the implementation reality is going to be sticky. Mainframes run the world’s most critical transactions—credit card swipes, airline bookings, government benefits. The idea of letting an "agentic" system—one that by definition creates its own path to a goal—have write access to these environments is terrifying for a risk-averse bank.
I am heartened to see they are focusing heavily on the "framework" aspect. You cannot just drop an agent into a mainframe; you need guardrails, or what we might call "digital trust." Kyndryl seems to have anticipated this by wrapping the agents in the governance layers of Kyndryl Bridge. It is a smart play. They are effectively saying, "We will give you the autonomy you need, but we will put it in a straitjacket of compliance first."
Still, the "black box" problem persists. If an agent optimizes a piece of JCL (Job Control Language) in a way that saves 30% CPU but introduces a subtle logic flaw that only triggers at month-end, who is responsible? The "human in the loop" concept is fine in theory, but if the human doesn't understand the complex code the agent just wrote (because of the aforementioned skills gap), that loop is broken.
Looking Ahead
Going forward, I am going to be watching the "trust curve" very closely. We are seeing a lot of noise in the market right now—McKinsey’s recent "State of AI in 2025" report notes that while over 60% of companies are experimenting with agents, a vast majority are stuck in pilot purgatory, unable to scale. This aligns with what I am observing in the field; everyone wants the result of agentic AI, but no one wants the liability.
Kyndryl has a distinct advantage here over competitors like DXC or Accenture. Because Kyndryl owns the managed services layer for so many of these mainframe shops, they are already the ones "turning the knobs." If they can prove that its agents are safe because they are the ones managing the risk, they might just crack the code.
However, do not expect overnight adoption. We will likely see a "trough of disillusionment" where early pilots struggle with governance. The winners will be the ones who treat these agents not as magic wands, but as junior employees who need rigorous supervision. I suspect we will see Kyndryl push hard on "Digital Trust" services in the coming quarters to soothe these fears. If they can convince a bank's risk officer that an agent is as reliable as a 30-year veteran systems programmer, they will have effectively cornered the market on legacy modernization. But that, my friend, is a very big "if."
Steven Dickens | CEO HyperFRAME Research
Regarded as a luminary at the intersection of technology and business transformation, Steven Dickens is the CEO and Principal Analyst at HyperFRAME Research.
Ranked consistently among the Top 10 Analysts by AR Insights and a contributor to Forbes, Steven's expert perspectives are sought after by tier one media outlets such as The Wall Street Journal and CNBC, and he is a regular on TV networks including the Schwab Network and Bloomberg.