Productivity as a desired AI outcome dropped 25.7 percentage points in a single year. Let that land for a moment. From 67.5% to 41.8%, according to the 2026 State of the CIO survey. Automation fell 14.9 points alongside it. Meanwhile, innovation and modernisation nearly doubled to 32.4%, and scalability climbed to 32.8%.
That’s not a gradual shift. That’s CIOs collectively deciding that “making things faster” isn’t the point anymore.
The question has changed. It used to be “Can AI make my team 10% more efficient?” Now it’s “Can AI help us build things that didn’t exist before?” And once you ask that second question, everything downstream changes — how you evaluate investments, where you deploy, what success actually looks like.
The efficiency argument for AI? Dead. What’s replacing it is far more interesting.
The efficiency ceiling nobody talks about
Here’s the uncomfortable truth about the AI productivity thesis. It worked — and then it stopped working. Early adopters saw genuine gains in document generation, data analysis, customer support, code development. Real gains. Measurable ones.
But the returns plateaued embarrassingly fast. The tasks most amenable to AI automation — summarisation, drafting, classification — turned out to be the tasks with the lowest strategic value. Think about it: an organisation that drafts emails 40% faster isn’t 40% more competitive. It’s marginally more efficient at a commodity activity. Nobody’s winning market share with faster email.
CIOs noticed. AI/ML spending more than doubled over 2025, jumping from 13.3% to 29.5% of IT budgets — the fastest-growing line item by a wide margin. But the composition of that spend flipped. Less on productivity tools. More on platform capabilities that enable entirely new ways of working.
The number that really matters, though: pilot-stage AI adoption collapsed 31.2 percentage points, from 68.5% to 37.3%. That’s the largest single-item swing in the entire survey. Three-quarters of CIOs — 74.2% — now report having thorough, well-formed AI implementation plans. The “let’s try AI and see what happens” era is definitively over.
What “previously impossible” actually means
Innovation over efficiency sounds great in a board presentation. But what does “previously impossible” look like when you strip away the buzzwords?
Start with autonomous process execution. Invoice processing, compliance monitoring, supplier qualification, employee onboarding — these aren’t being made faster. They’re being fundamentally restructured. End-to-end AI execution with humans stepping in at defined checkpoints, not at every step. That’s a different architecture entirely, not an upgrade to the existing one.
Then there’s cross-system intelligence — and this one genuinely changes the game. AI can now synthesise information across ERP, CRM, HR, supply chain, and financial planning simultaneously. A manufacturing CIO can ask “What’s the relationship between our workforce scheduling, production quality, and customer satisfaction?” and get an analytically grounded answer. That question wasn’t slow to answer before. It was impossible to answer. Full stop.
Adaptive operations round out the picture. Business processes that reconfigure themselves in real time — pricing that adjusts to market signals, staffing that responds to demand forecasts, procurement that reacts to supply chain disruptions before they hit production. The jump from reactive to predictive to adaptive represents a capability tier that static process automation could never reach.
Every one of these requires something that often gets overlooked: a process orchestration layer connecting AI capabilities to operational workflows. Without it, AI remains a clever tool sitting next to your business processes. With it, AI becomes an operating capability woven into them. Uniksystem’s low-code BPM platform fills exactly this gap — connecting AI-powered processes across enterprise systems with the governance, audit trail, and adaptability that production environments demand.
Multi-agent AI needs a control plane. Most organisations don’t have one.
The technology shift underneath this CIO pivot? Multi-agent AI. Gartner calls it a top 2026 trend, predicting that by 2027 a third of agentic AI implementations will combine agents with different specialised skills to tackle complex cross-domain tasks.
More than three-quarters of CIOs expect their organisations to have invested in agentic AI by year-end. Why the urgency? Because enterprise processes are inherently multi-domain. A single procurement process touches finance, compliance, supplier management, logistics, and legal. No lone AI agent handles all of that well. But give each domain its own specialist agent, put a process orchestration layer on top to manage the workflow and exceptions, and suddenly you’ve got something that actually works.
Here’s where the convergence of agentic AI and BPM becomes strategically decisive. BPM provides exactly what agentic AI lacks on its own — structure, governance, auditability, human-in-the-loop control. Without that orchestration layer, multi-agent systems become autonomous black boxes that nobody fully trusts. With it, they become arguably the most powerful automation capability enterprise technology has ever produced.
Gartner puts it plainly: the interaction model for enterprise operations is shifting to “prompt engineering, policy definition, and workflow orchestration.” The CIOs who’ve figured out that workflow orchestration is the control plane for agentic AI — not an afterthought bolted on later — are the ones building competitive moats that are genuinely hard to replicate.
Follow the money
This pivot isn’t philosophical. It’s showing up in budgets.
Some 56% of CIOs are expanding investment in business process and IT automation — the highest single category. Not chatbot deployment. Structural process redesign powered by AI. Another 55% are expanding security and risk management spending, because securing agentic AI is a fundamentally different problem than securing traditional applications. And 54% are growing data and analytics budgets, since innovation-grade AI needs integrated, real-time, governed data — the departmental data lakes that sufficed for productivity use cases simply won’t cut it.
The deeper insight: innovation through AI isn’t a new budget line. It’s a reconfiguration of how existing IT money gets spent. Less on point solutions, more on platforms. Less on task automation, more on process transformation. Organisations aren’t necessarily spending more in total. They’re spending very differently.
By 2028, Gartner predicts at least 15% of work decisions will be made autonomously by AI agents — up from essentially zero in 2024. The organisations building the orchestration and governance infrastructure right now will be ready for that world. The rest will face a capability gap that no amount of catch-up investment closes quickly.
The choice you’re actually making
If your AI strategy is still framed around “making things faster,” you’re optimising for a game that CIOs have already stopped playing. The question worth asking: what could your organisation do that it genuinely cannot do today?
If your AI investments don’t have a process orchestration layer underneath them, they’ll stall at pilot — no matter how impressive the demos look. Models without workflow integration are exactly that: demonstrations, not operational capabilities.
And if your agentic AI initiatives launched without governance — clear autonomy levels, oversight policies, decision accountability — you’re building risk faster than value. Every CIO succeeding with agentic AI in 2026 deployed governance before autonomy. Not after the first incident.
Get this wrong, and your AI programme becomes an expensive collection of disconnected pilots that never quite reach production. Get it right, and you’re building capabilities your competitors literally cannot replicate from a standing start.
Jorge Pereira / CEO | Uniksystem