CIOs Turn To Proportional Guardrails As Blanket AI Review Breaks Down

Enterprise AI governance is starting to look less like a single gate and more like a dial. Instead of applying one approval process to everything an employee or an agent might do, a growing number of technology leaders are setting the weight of oversight to match the size of the risk. Low-stakes work gets near-total freedom; the checks tighten only as the potential for damage grows. Calibrating control this way speeds the organization up rather than slowing it down, because most work no longer waits in line for approval. And it keeps the CIO at the center, as the leader who decides where the lines sit and who stays accountable when something crosses them.

Krishna Prasad is Chief Information Officer and Chief Strategy Officer at UST, where as strategy chief he sets the company's direction and drives its cross-functional growth initiatives, and as CIO he leads its internal digital and technology efforts. He's spent two decades at UST in leadership roles across the US and India, a stretch the company credits with more than 20x growth, and he holds an MBA from Ohio State's Fisher College of Business, where he serves on the Dean's Advisory Council. Before UST, he was an Associate Partner at IBM Business Consulting Services and a Principal Consultant at PwC, and he founded the consulting and staffing firm Global Conductor, building it into a multimillion-dollar business within two years. He describes his work as blending business, technology, and organizational acumen, an instinct that shapes how he thinks about giving teams room to move with AI.

"We can't police everything," Prasad said. "As long as the damage can be limited, we allow people to do what they want and be successful within that. But as it gets into a larger scope, that's where the need to pay attention increases." The restraint is deliberate. In his model, trying to review every action equally is both impossible and self-defeating, because it buries the few decisions that actually warrant scrutiny under thousands that don't.

• Autonomy scaled to impact: Prasad sorts the work into three bands. "Think of it as small, medium, large," he said. The small is individual-productivity work, where a person can drive their own autonomy because the damage is contained; the medium is what a team accomplishes together; the large is organizational, cutting across departments, and each band carries a different level of checks and balances to clear. The design goal, he added, is to promote more autonomy in everything that gets done while holding on to a clear sense of accountability. It's a view that treats earned autonomy as something sized to consequence rather than handed out or locked down across the board.

• Standards replace manual review: Sizing oversight that way only works if the low-stakes lane is genuinely fast, and the old change-management reflex can't deliver that. "Earlier you could say somebody is going to manually review these things and say it's okay to move forward," Prasad said. "Today you don't have the time, and even if you did, there's so much coming your way that it's almost impossible to stay on top of it." His answer inverts the model: the technology function defines the standards, and AI measures against them to surface the anomalies worth a human's attention. He's already building this into his own environment, pairing it with automated provisioning so teams can experiment safely and move to production when ready, a shift toward governing delegated authority rather than policing every request by hand.

• Business ownership decides the spend: The guardrails answer how; the business still has to answer why. Prasad is firm that AI investment can't live in a technology silo. "Technology cost should never have been thought of as something separate. It almost always requires business justification and ultimately business ownership to get the best value out of it," he said, and the discipline of tying spend to measurable return is, in his experience, what separates the initiatives that scale from the ones that stall. Investing in AI for its own sake, he noted, doesn't help anyone.

The tiered model earns its effort through what it does to the pace of the business. When the checks are proportional and the standards do the first pass, the work that used to queue for a weekly meeting simply moves. Prasad has watched that play out across his organization, and he's clear that the gains aren't theoretical. The numbers, he argues, are already on the table.

• The acceleration is measurable: "We can easily point to anywhere from 30 to 50% productivity improvement for developers who know what they're doing," he said. The effect compounds past individuals: systems that would once have taken a year have reached full production, concept to deployment, in three or four months. Cross-functional processes show it too, with vendor onboarding, which touches compliance and several departments, collapsing from weeks to a few days. None of it removes the work, but it does shorten the clock, the kind of change that surfaces once building stops being the constraint.

• Cost discipline without the clamp: Speed at that scale runs up a bill, and the bill now needs its own discipline. Gartner has forecast worldwide end-user spending on AI models and platforms will reach roughly $64 billion in 2026, with enterprise budgets facing new scrutiny around usage efficiency and cost control. Prasad won’t let that scrutiny choke off access. He wants basic AI capability in everyone’s hands, the way a CIO rolls out Excel or Word, then watches what happens through telemetry. “We’re also monitoring usage,” he said, “and once it goes beyond a certain level, we look at approval. Not necessarily to beat you up, but to understand what value you got, and how we can apply that into other functions.” The balance matters to him because both failure modes are real. “If you start putting controls very heavily upfront, it stifles innovation,” he said. “If you just let things completely loose, then it costs the business dearly.” Model choice follows the same instinct. Most of the business, in his view, runs fine a step behind the frontier, a call that benchmarks built for the business are better placed to make than generic leaderboards.

The hardest adjustment has nothing to do with the systems. It falls on who does the work and what the organization values in them. As AI commoditizes raw technical depth, the traits that used to sit lower on the list are rising, and that's reshaping the shape of the team itself. The pyramid Prasad once relied on is flattening from the bottom.

• The pyramid is flattening: For years UST, like most services firms, staffed a wide base of junior talent guided by a few seniors, a structure built to manage cost. "Now what we are finding is that it's probably not as effective," Prasad said. The base is thinning in favor of people who bring broader problem-solving judgment, because the traits AI can't supply—the judgment, the feel for quality, the willingness to own what ships—are the ones that now carry weight. The practical demand on a technologist has shifted accordingly: the developer who once got by knowing a single system now has to understand how the whole business creates value and makes money, since AI-era processes rarely stay inside one department, a reordering that puts domain understanding ahead of tooling.

Prasad frames all of it as a start rather than a settled state. "I'd think of this as the early stage of a fairly significant transformation that's going to be AI enabled," he said. His advice to peers walking the same road is plain: stay close to the business, deliver quick wins, and resist the urge to operate in a silo. The real discipline, in his framing, is using governance to decide how much speed the business can afford in each place, rather than treating it as a brake on all of it.