Enterprises pivot to smaller, measurable AI projects by 2026

Enterprise buyers are set to pull back from large-scale artificial intelligence deployments in favour of smaller, measurable projects in 2026, as anxiety over AI hype, security and data readiness reshapes spending plans across both private and public sectors.

Senior executives working with large organisations say the past year exposed a gap between experimentation with generative tools and the hard work of embedding AI agents into core workflows, prompting a shift towards incremental gains and more resilient architectures.

AI adoption

Jaimie Tilbrook, Chief Product Officer of data management firm EncompaaS, said many enterprises and government agencies still treat AI as an add-on rather than a change in how work gets done.

"It was surprising how many organizations were turning a blind eye to AI, either deciding not to use it at all or assuming they've adopted AI simply because staff can ask ChatGPT or Copilot questions. This misses the real opportunity: deploying AI agents to meaningfully improve workflows and outcomes. For commercial organizations, that means gaining competitive advantage. For government and not for profit, it means doing more with less, streamlining processes, reducing costs, and improving services for constituents," said Tilbrook, CPO, EncompaaS.

Large organisations have invested heavily in pilots built on public generative AI platforms. Many are now assessing how much of that experimentation translates into operational improvements, and what data foundations are necessary before wider deployment.

Incremental projects

Tilbrook expects boards and technology leaders to favour limited-scope AI initiatives with clear metrics over broad transformation programmes in the year ahead.

"A major disruption we'll see in 2026 is that companies will decide to move away from the hype of instant AI transformation and toward guaranteed incremental gains. Instead of trying to deploy a large-scale AI implementation with a high degree of uncertainty of success, companies will instead invest in smaller AI projects with a 100% success rate," said Tilbrook.

This approach aligns with a wider trend in corporate IT governance. Risk committees and regulators have raised questions about how organisations validate model performance, handle data lineage and address bias. Smaller projects offer a contained environment and more predictable returns in areas such as document classification, routing, and targeted process automation.

Enterprise software

At the same time, some large providers expect the economics of software buying in the cloud to change as AI becomes more embedded in daily operations.

Ramy Houssaini, Chief Cyber Solutions Officer at Cloudflare, said boards are starting to weigh resilience and flexibility more heavily than simple subscription cost when they select software models.

"The traditional SaaS model-defined by static features and centralized data silos-is nearing its end. Enterprises are now demanding AI-native, real-time, context-aware services. 2026 will accelerate the shift from "application consumption" to AI-as-a-Service. Organisations will prioritize deploying domain-tuned models at the edge, keeping sensitive data local, and paying for intelligence over software seats. While SaaS won't disappear, its dominance ends as AI agents become the primary interface for enterprise workflows," said Houssaini, Chief Cyber Solutions Officer, Cloudflare.

Vendors and customers have begun to test alternatives to seat-based licensing as usage patterns change. As AI agents handle more queries and transactions autonomously, software margins may hinge more on quality of insight and model performance than on user counts.

Licensing models

Houssaini said traditional subscriptions that tie fees to the number of staff using an application look less suited to an era of automated agents and continuous data flows.

"The old way of buying software (SaaS) meant paying a monthly fee for every employee ("seats") to use a set of fixed features, with all their data locked in a central silo. That model is breaking. In 2026, the focus will shift to AI-as-a-Service. Companies will demand software that is smart, real-time, and customized. They will pay for the actual intelligence and insights the AI provides, not the right to use the program. This move pushes smart AI assistants to the forefront and requires keeping sensitive data secure and close to home. While SaaS won't disappear, its dominance will end as AI agents become the primary interface for enterprise workflows," said Houssaini.

Many cloud providers already offer usage-based pricing for AI inference. Industry analysts expect experimentation with new metrics such as cost per decision, cost per task, or outcome-based contracts as buyers renegotiate long-term agreements.

Industrial automation

Houssaini said industrial environments will sit at the sharp end of this shift, as operators bring AI closer to machinery and critical infrastructure.

"Think of old factories and power plants (Operational Technology or OT) like a house alarm-they only react after something breaks. In 2026, Industrial AI changes that. AI models will constantly run the show, tuning machines and optimizing systems in real-time, moving from watching to driving operations. This sudden level of automation requires a huge security upgrade. Since you can't install security software on every single robot or sensor (the IoT devices), security needs to become invisible. We'll see a massive switch to a new security model called Agentless Zero Trust, which checks the identity of every machine interaction instantly and automatically, making the whole network fabric the trusted security guard for automated equipment," said Houssaini.

Manufacturers and utilities are under pressure to reduce downtime and energy use while meeting stricter cybersecurity rules for critical systems. That is leading to increased focus on network-level controls and identity checks for machine-to-machine communication.

Organisations across sectors now face a dual challenge. They must upgrade security architectures and data management while also recalibrating expectations around what AI can deliver in the near term.