IBM delivered its most comprehensive AI announcement in years at Think 2026. The vision is credible. The gap at the z/OS operational layer is real.
IBM Think 2026 happened in Boston on May 5–6. The announcements were real and the strategy was coherent. For the first time in several years, IBM had a single narrative running through the whole portfolio: build the AI operating model that lets enterprises actually capture value from their AI investments.
That narrative is correct. The 25% ROI figure that IBM’s own CEO study surfaced – only one in four AI initiatives delivering expected return – is an industry-wide problem that IBM is uniquely positioned to address. The governance layer, the orchestration layer, the enterprise trust model: all real, all needed, all things IBM can credibly sell.
But there is a gap. It sits at the z/OS operational layer, and Think 2026 did not fill it.
One disclosure: I am not an IBM employee. IBM does not pay me. I have been building mainframe software independently since 2004 and have no financial relationship with IBM.
IBM Bob – the AI-powered software development platform – is now generally available for modern and legacy environments including COBOL. Early users are reporting real productivity gains. This is no longer a research project. It is a product.
IBM Concert, the AI operations platform that correlates signals across applications, infrastructure, and network without forcing organizations to rip and replace existing tooling, is in public preview. The framing is right: cross-domain visibility, context-driven decisions, human oversight maintained. This is how operations teams actually want to work.
IBM SQL Data Insights Pro for Db2 z/OS – announced May 6 – embeds AI intelligence directly into Db2 query processing without requiring data movement. In-place. No separate pipeline. No latency penalty from moving data off the mainframe to run AI on it.
IBM Data Gate for Confluent synchronizes Db2, IMS, and VSAM data to Confluent Cloud via change data capture, enabling real-time fraud detection and analytics from mainframe data without impacting mainframe performance.
And IBM LinuxONE 5 is generally available with quantum-safe cryptography built in today – not a roadmap item. For regulated environments running financial transactions, this matters.
All of this is real. IBM is investing in the mainframe AI stack at every layer: development tools, data access, query intelligence, operational visibility, cryptographic security. The commitments are credible.
The most significant thing IBM said at Think 2026 about mainframe AI was not in a keynote. It was in the fine print.
IBM Bob’s Premium Package for Z – the mainframe-specific AI development capability that would bring Bob’s full intelligence to z/OS environments – is still in private technical preview. Not public preview. Not limited availability. Private technical preview.
This is IBM acknowledging, in product lifecycle language, that the mainframe AI problem is harder than the general enterprise AI problem. The same company that has GA’d a comprehensive AI development platform for modern and legacy systems cannot yet GA the z/OS edition.
This is not a criticism. It is a fact about the environment. z/OS application intelligence requires understanding SMF records, CICS transaction flows, Db2 catalog structures, JES job execution traces, and decades of operational context that does not exist in any training set. IBM has more of that context than anyone. And they are still in technical preview.
What that tells the market: z/OS-native AI capability is an open problem. IBM is working on it. It is not solved yet.
IBM Concert is the right answer to a real problem. The problem is cross-domain operational visibility – correlating signals from applications, infrastructure, and network into a unified view that enables coordinated response rather than siloed firefighting.
Concert builds on Instana, Turbonomic, and SevOne. All three are serious monitoring platforms. None of them go deep on z/OS internals.
When Concert surfaces a problem in the mainframe layer – an abend pattern, a CICS response time degradation, a batch job that is running long for the third consecutive night – it can tell you that something is wrong in the mainframe layer. It cannot tell you what. It cannot read a SYSOUT dump and identify the failing module. It cannot correlate a series of S0C7 abends with a recent COBOL compile and tell you which copybook changed. It cannot trace a CICS task that abended partway through a DB2 call and explain why the rollback did not complete as expected.
That is not a design failure in Concert. Concert was designed for cross-domain correlation, not z/OS-specific diagnostics. The two problems are genuinely different, and solving one does not solve the other.
Operations teams that deploy Concert will get better visibility across their hybrid estate. When the alert comes from z/OS, they will still need someone who can read the dump.
IBM zSecure Secret Manager is planned for June 2026. The announcement is real: enhanced security automation for RACF on z/OS, integrated with IBM Vault Self-Managed for IBM Z and LinuxONE, focused on certificate lifecycle management.
Certificate lifecycle is a legitimate and important problem. Certificates expire. Automated rotation is better than manual tracking. IBM is solving that problem.
What IBM zSecure is not doing: automated vulnerability scanning of the RACF environment itself. Not scanning for misconfigured datasets with universal access. Not flagging users with excessive authority that violates the principle of least privilege. Not identifying RACF controls that fail PCI DSS or ISO 27001 requirements. Not producing the compliance certificate that an external auditor needs to sign off on the mainframe security posture.
These are different problems from certificate lifecycle. The distinction matters because the buyer is different. The CISO buys certificate management. The external auditor – the Big 4 accounting firm running the annual security assessment, the cyber insurance underwriter deciding whether to issue the policy, the compliance consultancy certifying the bank for its next regulatory examination – needs something else entirely.
They need a tool that runs a structured assessment, maps findings to compliance frameworks, and produces a defensible, reproducible output they can put in a report. IBM zSecure does not produce that. IBM did not announce anything at Think 2026 that fills that gap.
Think 2026 is market validation, not market saturation.
IBM Concert confirming that hybrid cloud AI operations is a real category is market validation for the enterprise operations ecosystem. IBM SQL Data Insights Pro confirming that in-place AI on Db2 z/OS has commercial value is market validation for z/OS-native AI. The Bob Z package still being in technical preview confirms that the z/OS AI problem is genuinely hard – and genuinely open.
The organizations that should pay attention to the gaps are the ones that will be asked “what happened and why” when a batch run fails, and the ones that will be asked “can you certify the mainframe security posture” by their auditors this quarter. Concert cannot answer the first question. zSecure cannot answer the second one.
IBM is building the enterprise AI operating layer. It is doing it well. The z/OS operational diagnostic and security intelligence that needs to plug into that layer is still an open problem. The gap is real. The market has not filled it yet.
IBM Think 2026 confirmed that the mainframe AI market is real and IBM is investing in it seriously. That is the right read of the announcements.
It also confirmed, in the fine print, that z/OS-native operational AI is an unsolved problem – the Bob Z package in private technical preview is IBM saying so in product lifecycle language.
Concert tells you something is wrong. It cannot tell you what. zSecure manages certificates. It does not certify compliance. Those gaps are still open.
Also worth reading: The Batch Window Does Not Follow the Workload · People Can Sleep Peacefully With Mainframe · So You Decided Mainframe Is Cheaper Than Broadcom. Now What?