Gartner’s H2 2026 CIO Report: Why AI Operating Models Matter for Making AI Work 

My response to Gartner’s H1 CIO Report focused on what companies should establish before deploying artificial intelligence (AI), and honestly most of the mid-market still isn’t there yet. Based on the Chief Information Officer (CIO) questions and findings Gartner presents for their H2 report, the focus is now on adaptation, AI resilience and the organizational changes required to make AI work at scale. The report addresses distributed decision-making, resilience, application modernization and human judgment. If you don’t get those things right, a team can gain a useful capability and still struggle to put it to work because its authority, data or supporting processes have stayed the same. In short, a CIO can approve an AI tool faster than the company can change the way it makes decisions and executes.

My (maybe controversial) take: H2 reads like a warning about what happens when we don’t stop to put the right foundations in place, but now we’re accelerating processes that don’t have the right governance and organizational structure around them. Didn’t listen to H1 warnings? Do not pass GO. Do not collect AI ROI.

Outdated Models and the Need for a Modern AI Operating Model – It’s not just IT

Gartner reports that 85% of information technology (IT) leaders plan to deliver AI-enabled business services in the next three years, while 50% of CIOs agree their IT function cannot keep pace with change. In fact, Gartner forecasts that, by 2028, 50% of enterprises will fail to realize measurable AI business value because of outdated IT operating models. This highlights the growing importance of establishing an effective AI operating model that can adapt alongside business needs. The report also says only 24% of CIOs consider themselves highly effective at establishing IT operating models that adapt to business needs. 

Is this an IT problem? I don’t think we should be so quick to throw IT under the bus. It isn’t the only place where outdated operating models are found. See How Leadership Direction Drives AI Adoption for more details.

How do we make decisions in this new uncertain environment? How do we move quickly without ballooning risk and creating chaos? What organizational structures do we need to put into place so we learn from failures and use them as steppingstones for the next success? Surely, these are not “IT” questions.

It seems to me that the whole organization needs to take a step back and question current operating models – and this needs to come from the CEO, not IT. Whatever happens, there’s no doubt CIOs need help. Either IT operating models will adapt to business needs or IT functions will end up distributed across the organization.

Decision Authority Across Teams

One of the recommendations in the Gartner report is distributing decision-making capability with guardrails across the organization. A successful AI operating model depends on empowering teams, maintaining appropriate governance through a clearly defined AI governance framework and judging architecture by strategic outcomes. In practice, that means specifying which choices a team can make, what data it can rely on and when an idea needs broader review. If every exception still goes to the same IT approver (a common model), that person’s availability limits how quickly exceptions are resolved, even after AI speeds up individual tasks.

The important phrase there is “with guardrails”. What those guardrails are should have already been decided by the organization’s leadership in consultation with IT, legal, risk management and compliance.

You may want to take a look at the governance and guardrails article from Withum’s AI Readiness portal for some ideas.

AI Resilience in Day-to-Day Operations

Next, the report recommends attention to vendor dependency, modular architecture, security, continuity and workforce agility. Together, these capabilities form the foundation of long-term AI resilience and support an AI governance framework that can adapt as technologies, vendors and business needs change. Teams are encouraged to establish how work continues during an interruption, who can respond and whether they can change components independently, then include the cost of maintaining those options.

Withum’s vendor-agnostic approach has led us to the same conclusion: if we build for a client, we want to ensure that models can be unplugged and replaced with no or limited rework. Things are still too volatile to put all your eggs in one model basket.

Modernization Costs and Continuity

Gartner forecasts that generative AI will reduce application-modernization costs by 30% by 2028. My take is that this means leaders can revisit projects previously rejected on cost grounds, while following Gartner’s advice on testing and maintaining operations during the change.

Withum has already helped clients do exactly this. AI-assisted application modernization is more than hype. It is already changing the game – so much so that we wonder if the 30% reduction by 2028 isn’t too conservative. For SMBs and the mid-market, we think the benefits will be more immediate. Learn more: A New Agile Approach to Application Modernization Services and What to Expect.

Judgment and Skills in the Work Itself

Gartner reports that 51% of CIOs see the gap between evolving AI skill requirements and available talent as their primary challenge. It recommends combining human context and judgment with AI speed and embedding learning in the work. Reviewing exceptions with experienced colleagues allows employees practice recognizing when AI output is deficient.

Consider this example: a customer-service employee reviews an AI-drafted reply about a failed repair. The response is clear, polite and consistent with policy, but it misses that the repair has failed twice, and a colleague already promised a different remedy. The employee needs access to that history and enough authority to arrange or escalate the remedy. Sending another reply without addressing the earlier promise could leave the customer contacting the company again and eventually taking their business elsewhere.

“Training for judgment” is something we’ve been talking about for the last year and a half. Take a look here to see what judgment training actually looks like in AI decision-making. Not only do line employees need to learn how to evaluate AI outputs, but managers have to relearn the review process; the things we looked for when interns or junior staff gave us something to review – spelling, grammar, lack of proper segues between concepts – can now look polished even when the substance is wrong. Now we need to review the sources, check whether the answer fits the situation and watch for language that sounds polished but lacks substance.

Final Word

My H1 response focused on what needed to be in place before deploying AI. The adaptation, AI resilience and AI operating model questions raised in the report make those unfinished decisions harder to avoid. I don’t think CEOs can leave this with the CIO and expect the rest of the organization to catch up. Resilience and adaptation are vital, but what are they built upon?

Remember: no matter how fast you run, it isn’t a home run if you don’t touch first base.