From spectacle to systems: Why 2026 may force AI to grow up

Microsoft chief executive Satya Nadella is not predicting an artificial intelligence (AI) breakthrough so much as an inflection point. In a recent blog post, he argued that 2026 will mark a decisive shift for AI. He predicted that it will move from a phase dominated by discovery, demonstrations and rapid model development to one defined by diffusion, where the technology’s value will be judged by how deeply and reliably it works in the real world.

For an industry accustomed to celebrating scale, speed and benchmarks, Nadella’s framing is deliberately sobering. He suggests that AI is currently running ahead of itself, with capabilities advancing faster than their ability to generate tangible outcomes for businesses and society. The next phase, he argues, will be less about marveling at what models can do and more about proving what they are actually good for.

At the heart of Nadella’s argument is what he describes as a “model overhang”. AI systems are becoming more powerful at a breathtaking pace, but the infrastructure, workflows and human practices needed to deploy them effectively are lagging behind.

“The pace of AI development has been extremely fast, but admitted that the ability to use these powerful systems in practical ways has not kept up,” Nadella acknowledged in the blog post. The result, he said, is a growing disconnect between technical progress and real-world value creation.

This is not unusual in the history of technology. Breakthroughs often arrive before markets, organisations and cultures are ready to absorb them. Nadella likens the current moment to the early stages of a marathon rather than a sprint. “We are still in the opening miles of a marathon,” he wrote, adding that while progress has been remarkable, much about AI’s trajectory remains uncertain.

Crucially, Nadella warns against mistaking demos for deployment. “Many of today’s AI capabilities are yet to translate into tangible outcomes that improve productivity, decision-making or human well-being at scale,” he said. That distinction—between what is possible in theory and what works reliably in practice—is likely to define how AI investments are evaluated over the next few years.

From ‘bicycles for the mind’ to cognitive tools

Nadella revisits a familiar metaphor from the early days of personal computing to explain what needs to change. Steve Jobs famously described computers as “bicycles for the mind”, tools that amplified human capability rather than replaced it. Nadella argues that this idea must evolve in the AI era.

“This idea needs to evolve in the age of AI,” he noted. “Instead of replacing human thinking, AI systems should be designed to support and strengthen it, acting as cognitive tools that help people achieve their goals more effectively.”

This reframing pushes back against binary debates that often dominate AI discourse—automation versus augmentation, replacement versus enhancement. For Nadella, the more useful question is not how advanced a model is, but how people actually work with it.

According to him, “the real value of AI does not lie in how powerful a model is, but in how people choose to use it.” That shifts the focus from technical prowess to human adaptation: how teams integrate AI into decision-making, creativity and problem-solving without surrendering agency or accountability.

He also cautions against simplistic judgements of AI outputs as either crude or sophisticated. The more meaningful measure, he suggests, is whether humans can use these tools to operate more effectively in everyday interactions, from work processes to societal systems.

Why systems, not standalone models, will matter

If the first wave of AI was about building models, Nadella believes the next wave must be about building systems. Advanced models, on their own, are not enough to deliver sustained impact.

He points out that despite exponential gains, AI still has “jagged edges”—limitations around reliability, context, memory and safety that make real-world deployment complex. Addressing these gaps requires engineering beyond the model layer.

The next phase, Nadella argues, involves constructing full AI systems that allow multiple models and agents to work together, retain memory over time, manage entitlements and use tools safely. This architectural shift, he says, is essential if AI is to move from impressive experiments to dependable infrastructure.

For businesses, this has practical implications. It suggests higher upfront complexity, longer integration cycles and a greater emphasis on governance and design. It also implies that competitive advantage may come less from access to the most advanced model and more from how well organisations build systems around it.

Beyond engineering, Nadella emphasises the importance of restraint. Not every problem needs AI, and not every application justifies the energy, computing power and talent required to run it.

He stresses that societal acceptance of AI will depend on its ability to demonstrate real-world results, particularly in areas that affect people and the planet. Decisions about where to deploy AI, he argues, will increasingly be moral and economic choices rather than purely technical ones.

Those choices are constrained by reality. Energy, compute and skilled talent are finite resources. Nadella notes that deciding how to allocate them will require broad consensus, not just within companies but across industries and governments.

Progress, he adds, should ultimately be measured by outcomes for individuals and society, not by abstract performance metrics. As with previous technology waves, the path ahead will be iterative, uneven and shaped as much by social acceptance as by technical possibility.

Marketing’s shift from tools to agents

Nadella’s cautionary tone finds echoes in how AI is already being discussed within marketing and advertising circles. While AI has rapidly become embedded in creative optimisation, audience targeting and bidding, industry leaders increasingly see its next phase as more autonomous—and more consequential.

By 2026, some expect AI in marketing to evolve from a tool that executes tasks to an agent that makes decisions. Today, AI systems tend to optimise in silos—creatives here, audiences there, bids somewhere else. Over the next 12 months, marketers anticipate agentic systems that understand business objectives, customer behaviour and brand strategy, then autonomously plan, test, spend and optimise across channels in real time.

For a BFSI brand like Kotak Mahindra Bank, this shift has concrete implications. Kedarswamy Ravangave, executive vice-president for marketing, said in an earlier interview with Campaign that such systems could deliver faster learning cycles, reduce wastage and enable far more personalised, compliant and context-aware customer journeys at scale.

That promise, however, hinges on exactly the kind of system-building Nadella describes—integrated, governed and aligned with human oversight rather than operating as black boxes.

Efficiency, economics and the cost question

Another recurring theme is efficiency. As AI capabilities expand, so do expectations that they will simplify complexity rather than add to it.

Latish Nair, chief digital officer for e-commerce at WPP Media, pointed to this potential in areas like measurement and modelling. “It currently solves for a lot of complexity as we’ve seen with WPP Open,” he said. “It will also give us an opportunity to convert customer occasions better.”

Nair highlights marketing mix modelling (MMM) as an example. Traditionally expensive and time-consuming, MMM could become more accessible if AI reduces the cost and effort involved. “I’m making a wish for: An MMM (meta mix modelling) of offsite versus onsite [media] to be done via AI. You just upload the data points and get a clear cutout. MMM is a very expensive proposition to do, so if AI could help to bring the cost down, that would be great,” he added.

This focus on economics reinforces Nadella’s argument: AI’s future will be judged less by what it can theoretically achieve and more by whether it delivers efficiency, clarity and value at scale.

A turning point, not a finish line

The sense that AI is approaching a turning point is not limited to corporate leaders. Scientist Geoffrey Hinton, often referred to as the ‘Godfather of AI’, recently said that the technology is advancing faster than he expected, particularly in reasoning and task completion. According to him, 2025 marked a major turning point, with systems becoming significantly more capable in the coming year.

Enterprises, meanwhile, are already moving agentic systems from pilots to production, even as they grapple with questions of trust, governance and return on investment.

Nadella’s message cuts through the hype with a reminder that maturity, not momentum, will define the next chapter. If 2026 does become the year AI moves from spectacle to substance, it will be because organisations learned to slow down in the right places—building systems, choosing applications carefully and measuring success by outcomes rather than headlines.

For industries like advertising and marketing, where experimentation is prized but accountability is non-negotiable, that shift may be less about adopting more AI and more about deciding what kind of AI is actually worth adopting.