{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreialv6t5gjxe353x2bvsbtbyjkukeqayug2oaoy4owjqam4bn3evqi",
    "uri": "at://did:plc:qz6ohvpdsdvv5kniizyfz25y/app.bsky.feed.post/3me5jy53kuwn2"
  },
  "coverImage": {
    "$type": "blob",
    "ref": {
      "$link": "bafkreif2gvr7q4mxdyurhpqa55ospfa2urhdhko5iq3ipqavlpuhu7i6tu"
    },
    "mimeType": "image/jpeg",
    "size": 14606537
  },
  "path": "/article/3630070/12-ai-predictions-for-2025.html",
  "publishedAt": "2026-02-05T10:00:00.000Z",
  "site": "https://www.cio.com",
  "tags": [
    "Artificial Intelligence, CIO, Digital Transformation, Emerging Technology, Generative AI, Industry, IT Leadership, IT Skills and Training, IT Strategy, Markets, Technology Industry, Vendor Management, Vendors and Providers",
    "According to Gartner",
    "Forrester estimates",
    "the CIO’s job",
    "unsanctioned AI tools",
    "Deloitte says",
    "small language models",
    "PwC",
    "tech-enabled governance approaches",
    "Forrester",
    "Sonar survey"
  ],
  "textContent": "Generative AI continued its breakneck pace of adoption and technical progress last year. Now, after so much trialing and tinkering, 2026 is shaping up as the year AI starts to deliver on its promise, and reveal its limitations.\n\nAccording to Gartner, enterprises will spend a staggering $2.5 trillion on AI this year, up 44% from 2025. But as enterprises demand to see ROI, Forrester estimates 25% of planned AI spend may be deferred into 2027. Here are 10 expert predictions for how AI will progress and expand in 2026, and where it’ll run into obstacles and delays.\n\n## From pilots to production\n\nThere was a lot of experimentation in 2025 due to the rapid evolution of AI technology. Agentic systems, MCP servers, increasingly intelligent models and falling inference costs all combined to expand the universe of AI use cases. Now it’s time to start putting all this learning into practice.\n\n“After years of experimentation, companies will need to be done with pilots and ready to move on to real AI transformation,” says Neil Dhar, global managing partner at IBM Consulting. “The proof now will come not from what AI can do, but from how to make AI deliver measurable results.”\n\n## Focus on ROI\n\nAs part of this maturing approach to AI technology, enterprises will also expect to see financial results of all their investments. “Real ROI will come from doing the hard miles, redesigning workflows, turning data into meaningful intelligence, and rethinking operating models to drive real impact,” adds Dhar. Plus, companies will need to ensure that AI spending is strategic and well controlled.\n\n“What starts as a small science project can balloon across an organization very quickly, and the costs balloon right along with it,” says Chris Bennett, VP of the global AI practice at Unisys. “The bigger issue underneath all that is alignment. If the business isn’t aligned with what the AI team is being asked to deliver, nothing moves forward.”\n\n## CIOs will become chief orchestration officers\n\nThe shift to agentic AI will redefine what it means to be a CIO. “My role as CIO is shifting from running AI experiments to orchestrating and governing outcomes for the enterprise,” says Shannon Bell, CIO and CDO at OpenText. “In 2026, I believe the CIO’s job will be focused on secure information management, coordinating how digital and human workforces function together, and making sure AI delivers measurable business impact at scale.”\n\nSo traditional IT management will need to adapt and evolve. “Managing an agent workforce differs from managing people,” adds John Cannava, CIO at Ping Identity. “It’s less about culture and more about governance and accountability. CIOs must define who has the authority to create agents, establish technical change controls, and ensure every agent’s work is tested, monitored, and trusted.”\n\nAnd the scope of this transformation is enterprise-wide as CIOs will move from managing systems to orchestrating how work flows across the business, says Saket Srivastava, CIO at work management platform Asana.\n\n## Shadow agents will create a security crisis worse than shadow IT\n\nIn a January survey by cybersecurity firm BlackFog, 60% of 2,000 employees at large companies say unsanctioned AI tools are worth the security risks if it helps to work faster or meet deadlines. Additionally, 21% believe their employer would turn a blind eye to the use of unapproved AI tools as long as work is completed on time. Agentic AI makes these tools even more attractive to employees, and riskier to enterprises.\n\n“Expect a surge in employees using unauthorized AI assistants or plugins to automate tasks,” says David Higgins, senior director of the field technology office at CyberArk. “This shadow AI phenomenon will mirror the earlier rise of shadow IT, creating new blind spots for security teams.”\n\nAnd shadow agents will accelerate data exposure faster than it can be detected, says Suja Viswesan, IBM security software leader. “As autonomous AI agents begin to operate independently across enterprise environments, often outside sanctioned workflows, they access sensitive data with minimal human oversight,” she says. That creates a new exposure problem. “Businesses will know data was exposed but won’t know which agents moved it, where it went, or why,” she adds.\n\n## From training to inference\n\nDeloitte says inference workloads will account for two-thirds of all AI compute in 2026, up from half in 2025. “The models have become smart enough that most organizations won’t need to train their own,” says Andrew Hillier, CTO and co-founder of Kubex, an infrastructure optimization firm.\n\nCompanies can use off-the-shelf models, and pay more attention to the inference side of the equation, he says, which changes dynamics about GPU availability, efficiency, and performance.\n\n“Considerations like latency, cost, and reliability are becoming more critical than raw compute power,” adds Barry Baker, COO and GM of IBM Infrastructure.\n\nThat will change how enterprises think of computing infrastructure, and allow for new capabilities while saving money. “Purpose-built AI infrastructure is already enabling applications that were impossible on generic platforms, such as real-time fraud detection or ultra-low latency decisioning,” he says. “And there’s sustainability. These fit-for-purpose approaches consume far less energy, which lower costs and carbon footprints. “\n\n## Domain-specific models will gain ground\n\nRecent research from Stanford shows that small language models are already very capable, inexpensive, and efficient. Plus, they can be run locally. As demand for AI grows faster than centralized data centers can support, small and domain-specific models can offer a way forward.\n\n“The age of massive, general-purpose models is giving way to smaller, more focused systems,” says Marlene Wolfgruber, AI lead and computational linguist at ABBYY. And these smaller models can be built around the language, logic, and compliance needs of specific industries or use cases.\n\n“Models trained on domain-rich datasets and tuned to the nuances of the processes within finance, healthcare, law, insurance, and other industries will outperform generic LLMs on accuracy, trust, and relevance,” she says. “Their value lies not in scale, but in specialization. 2026 will be the year enterprises realize that the best model isn’t the biggest — it’s the one that understands their problem.”\n\n## AI as UI\n\nWhen the internet emerged, applications moved from desktops to websites, and the browser became a dominant user interface. With smartphones, touchscreens became the new way for users to interact with apps. Now with AI, we’re moving to an era of natural language interactions with our tools via chatbots and other AI-powered options.\n\n“The prompt becomes the new UI for daily work,” says Bryan Wise, CIO at 6sense. “Instead of navigating dashboards, inboxes, and apps, employees will simply ask the system what they need to know, and AI will deliver answers, actions, and insights on demand.” But just because they’re using chatbots doesn’t mean users will be limited to just text-based interactions.\n\n## Governance moves to the core\n\nPwC says responsible AI boosts ROI and efficiency, but turning responsible AI principles into operational processes is a different story. That could change in 2026, though, with new, tech-enabled governance approaches.\n\nAnd Forrester further predicts that half of enterprise ERP vendors will launch autonomous governance modules this year, combining explainable AI with audit trails and real-time compliance monitoring.\n\n“In 2026, compliance will be coded directly into generative workflows, making governance an integral part of system design rather than an afterthought,” says Karthik Rau, CEO at Contentful. “What once slowed innovation will now become a foundation for scale, turning governance from a blocker into an architectural feature.” And it’s not just regulators that’ll want to see more responsible use of AI.\n\n“Customers will reward companies that can clearly explain how their AI system works, what data it uses, and why it makes certain decisions,” says Lisa Owings, chief privacy officer at Zoom. “Transparency will be the biggest differentiator among AI-driven companies, rather than just a compliance requirement.”\n\n## Vibe coding takes over\n\nVibe coding was coined by Andrej Karpathy in early 2025 and it went viral. It basically refers to a user providing high-level instructions and feedback, and the AI doing all the actual coding work. “Vibe coding may sound informal, leading some leaders to overlook it,” says Vikas Agarwal, PwC’s global and US advisory commercial tech and innovation officer. “In reality, it reflects a meaningful shift in how enterprises learn, decide, and build with AI.”\n\nIn a Sonar survey of over 1,000 developers released last month, 72% of developers who’ve tried AI coding tools now use them daily. In addition, 42% of code was AI-generated or assisted in 2025, and respondents expect that to go up to 55% this year, and 65% in 2027.\n\n## Enterprise software faces major shakeup\n\nAccording to Shannon Bell, CIO and CDO of OpenText, AI will replace traditional tools. “For me and for our teams, that means fewer standalone service desks, reporting, and workflow platforms, and more AI built directly into how work actually gets done,” she says.\n\nThe real value isn’t in reducing headcount, she adds, it’s about reducing tool sprawl and simplifying the environment. And as AI models and agentic systems get more capable, they’ll be able to handle tasks previously done by software products.\n\n“Budgets will shift from legacy cloud software to agentic automation,” says Andie Dovgan, chief growth officer of Creatio, an agentic AI firm. “Enterprises will decrease spending on traditional SaaS subscriptions and redirect those funds to agentic automation ecosystems.”\n\nThe way that enterprises buy software will also change, he adds, moving from paying for software seats to paying for autonomous outcomes.",
  "title": "10 AI predictions for 2026"
}