Over the past couple of months, improvements to long-standing IT automation platforms will start linking agentic AI directly to core enterprise systems — including ERP platforms and mainframes.
For years, businesses have relied on workload automation and orchestration tools, which originated long before cloud computing and DevOps became mainstream. These tools have since been updated to support newer trends and, more recently, have been enhanced to work with generative AI and agentic AI. Platforms in this space automate routine tasks and integrate varied, complex software systems into unified business processes with strong dependability, according to Dan Twing, an analyst at Enterprise Management Associates (EMA).
“Workload automation acts as the connective tissue as you transition between different application domains, keeping a process intact across multiple environments,” Twing explained. “Every significant organization leverages it in some form. … The cloud’s expansion over the past two decades would not have happened without that kind of support and automation underneath.”
On April 8, Broadcom rolled out Version 26 of its Automic software, which introduces a new AI Agent Job type. This enhancement allows the workload automation platform to function as a Model Context Protocol (MCP) server, bridging traditional IT orchestration with AI agents. Automic, along with other tools within Broadcom’s Agile Operations Division, can feed critical application data from ERP, mainframe, and core banking systems into Broadcom’s dedicated AI infrastructure suite built on VMware.
BMC’s rival to Automic, Control-M, released an AI assistant and a workflow creation tool on March 18 and is collaborating with early-adopter customers on workload automation driven by AI agents. On April 8, BMC published a roadmap commitment for AI agent support within its Automated Mainframe Intelligence (AMI) offering and broadened its AI-generated zAdviser Enterprise mainframe reports to also encompass distributed systems applications.
Deeper within the legacy infrastructure tier, IBM revealed a partnership with chip designer Arm on April 2, which will enable cloud and mobile applications running on energy-efficient processors to be ported into IBM Z and LinuxOne environments through virtualization.
Workload automation is what makes enterprises backwardly compatible. Dan Twing, Analyst, EMA
The common thread among all these upgrades is that tools enterprises have relied on for decades are being framed as a proven method for introducing deterministic orchestration and governance — bolstering the dependability and security of inherently unpredictable AI workflows. With that positioning, these vendors join a crowded field of IT companies making comparable arguments to enterprises. However, as Twing noted, these particular vendors already have deep access to mission-critical customer systems, which could be especially persuasive to large organizations.
“Workload automation bridges legacy and modern environments and allows them to coexist — and it isn’t simply one old world and one new world. There are about 15 layers of different eras stacked on top of each other,” Twing said. “Client-server systems are still operating out there, along with early cloud architectures. Workload automation is what makes enterprises backwardly compatible.”
Broadcom marshals workload acquisitions
Broadcom’s Automic is centered around Jobs — software elements that carry out commands across various environments, including operating systems, databases, enterprise applications like SAP and Oracle’s E-Business Suite, file transfers, and web services. Version 26 introduces an AI Agent Job type that manages AI agents and folds them into established workflows. This new Job type also extends Automic’s role-based access controls, logging, and audit capabilities to AI agents’ operations for proper governance. Version 26 also includes a natural language-powered workload execution interface built on a Python-based Code Assist tool for assembling data pipelines.
Rajeev Kumar
“A user can enter a simple prompt, and whichever large language model they’ve connected to Automic, combined with the product’s built-in configurations and grounding rules, will produce a workflow plan,” said Rajeev Kumar, head of products for workload automation at Broadcom.
For instance, a business analyst may want to pull data from Salesforce each morning at 6 a.m., transfer that data into BigQuery for analysis, then send the results to Looker for report generation, create an AI-generated summary of the report, and email it to their CEO. Automic can now draft a workflow plan for that process, Kumar explained, identifying the specific Jobs and workflow elements needed to accomplish the goal, presenting the plan to the user for review, and then deploying it once approved.
“AI is assisting software engineers,” Kumar said. “Our focus is on non-software professionals — the business analysts who have historically built these workflows but have had to lean on makeshift tools.”
Over the past decade, Broadcom has assembled a wide-ranging portfolio of hardware and software businesses, positioning it as one of the most important enterprise AI infrastructure vendors, according to Stephen Elliot, an analyst at IDC.
“Many people don’t grasp just how much internet traffic flows over Broadcom hardware,” Elliot said. “VMware is only one segment of that large infrastructure software division, and the Symantec acquisition, the CA acquisition, network software products, and chip software all play a role too.”
BMC takes cautious, strategic view
In its March update, BMC’s Control-M added compatibility for AI agent platforms from partners such as CrewAI, LangGraph, and Snowflake Cortex, along with a Jett AI advisor and workflow creation tool. Its multi-AI agent orchestration capabilities are still under development, said BMC CTO Ram Chakravarti.
“We’re tackling this in two stages: within the core product, you can now invoke individual agents through prebuilt integrations and embed them into workflows,” Chakravarti said. “At the same time, we’re co-developing with some of our top customers on high-impact use cases where custom agents are being orchestrated alongside the Control-M core, or even supplementary features like Managed File Transfer for federated data exchange with AI.”
If your AI use cases are not aligned with your broader digital business strategy, your AI pilots will stall as science experiments. Ram ChakravartiCTO, BMC
Federated data exchange is a mechanism through which querying tools access sensitive data stored within a partner’s infrastructure — or vice versa — without the data ever leaving a company’s network. It can be a critical step when initiating work with a new partner. Chakravarti said that one Control-M pilot customer has used AI agents to shorten the federated data exchange process from 30 days to under 12 hours. He declined to identify the customer or share the company’s size, other than to describe it as an “extremely large” organization.
“We already built many of the building blocks, and we continue to expand them, but for us, the priority is how Control-M can deliver predictable, dependable outcomes while handling dependencies
management, reliable handoffs, SLA adherence and a whole bunch of complexity management, as we do traditionally,” Chakravarti said. “Unless your AI use cases are aligned to your overall digital business strategy, your AI pilots are going to languish as science experiments.”
Dan Twing
BMC has also undergone extensive portfolio rationalization over the years, most notably by splitting its IT service management and operations management business, and its workload and mainframe automation business, into separate companies last year.
“Broadcom still has to extend [agentic AI support] to some of its other products, but created the foundation based on years of work, architecturally, to rationalize these things,” Twing said. “BMC didn’t have that challenge — it had different challenges, like going through SaaS modernization, while Broadcom added SaaS two years ago.”
AI expands mainframe modernization opportunities
Broadcom has begun integrating generative and agentic AI into mainframe management by adding MCP servers to its Rally Agile development and Endevor change management software that support mainframes alongside distributed systems. Broadcom also supports the open source Zowe framework for hybrid cloud mainframe integration, including Zowe MCP server. Finally, IBM’s WatchTower observability tool includes AIOps features for mainframes.
Updates to BMC’s AMI tool in April included enterprise application analysis reporting for its AI-driven zAdviser development productivity monitoring tool. The existing AMI AI assistant added integrations with the mainframe Knowledge Hub and a Knowledge Expert chat to pull in information from sources such as runbooks, tickets, log files and prior incident resolutions.
BMC’s statement of direction for AMI will move it beyond explanation and recommendations to autonomous AI agent-driven workflows for system and performance diagnostics, development workflows, security validation and operational recovery, learning from past incidents.
With this statement of direction, BMC is taking a more holistic and thoughtful approach to AI for mainframe modernization than Broadcom, said Steven Dickens, CEO at HyperFrame Research.
Steven Dickens
“Broadcom put an MCP server on the mainframe and then connected it to a bunch of legacy apps, so you can sort of interrogate them via an MCP server, which seems to me like table stakes, rather than a holistic deployment of AI,” Dickens said. “BMC is looking at, ‘How do we ingest support data? How do we ingest Redbooks and support databases and knowledge bases? How do we then do code explanation and automate ops with a broader radius of thinking?'”
IBM, Arm and history repeating?
In Dickens’ view, BMC has the most ambitious mainframe strategy, but IBM also has its Concert AIOps software that supports System Z automation, along with control over mainframe hardware, which it brought to bear in its recent agreement to support Arm chips.
IBM undertook a similar effort to integrate x86 chips into its zBX systems more than a decade ago, which can now support most major enterprise workloads but has a few known issues in areas such as storage resource management and, in some cases, third-party application support, Dickens said.
Opening the platform further to Arm chips could offer another avenue for third-party app compatibility, and there are strong incentives on both sides to make the Arm integration work, Dickens said.
“Whatever anybody says about the mainframe, it’s highly available, highly resilient, highly performant — it’s the fastest commercially available processor,” he said. “Arm is getting access to that instruction set collaboration with hundreds of chip developers and architects — IBM has got some chops in this space.”
However, Dickens said he doesn’t expect to see any shippable results from the collaboration until the launch of the next-generation System Z, likely in 2028, according to IBM’s typical three-year System Z release cadence. The latest z17 systems were launched in April 2025.
On the workload automation side, an October 2025 EMA Radar Report for Workload Automation and Orchestration placed IBM’s Workload Automation in the “strong value” category. This ranked below Control-M and Automic, which were among the tools in EMA’s top “Value Leader” category, alongside Stonebranch, HCLSoftware, Beta Systems and Redwood.
But overall, IBM and Red Hat have a strong set of hybrid cloud agentic AI tools to compete with, Dickens said, for enterprise buyers.
“When you look at Red Hat, you look at the OpenShift integration that they’ve done on the mainframe — IBM is not just having a mainframe tools conversation, it’s having a more holistic hybrid IT kind of conversation,” Dickens said.
Beth Pariseau, senior news writer for Informa TechTarget, is an award-winning veteran of IT journalism. Have a tip? Email her or connect on LinkedIn.
Dan Twing
Steven Dickens