Applied AI in 2026: Why Adoption Is No Longer the Hard Part

Three years ago, the question in most boardrooms was: "Should we use AI?" That question has largely been answered. The more important question now is whether the AI operating inside an organisation is creating measurable value or simply generating more activity.

McKinsey's latest global AI research found that 88% of organisations now use AI regularly in at least one business function. Generative AI use has risen sharply too, from 33% in 2023 to 79% in 2025. AI adoption is no longer a meaningful differentiator. It's becoming a basic requirement for staying competitive.

What separates organisations in 2026 isn't whether they have access to AI. It's whether they can turn experimentation into secure, reliable, measurable production systems. That's the challenge Applied AI is built to address.

Why we're doubling down on Applied AI

We didn't arrive at this focus by chasing a trend. Across our Google Cloud and AWS engagements with growth-stage companies in the Baltics and Ukraine, we kept running into the same pattern: teams that already had the data, already had access to capable models, and still couldn't get an AI initiative to run reliably past the pilot stage.

That's the gap we built our practice around. Not selling access to a model, since every credible vendor can already do that, but doing the less visible engineering work underneath it: identity and access controls that can safely let an agent act, data pipelines clean and governed enough to trust an automated decision, and security reviews built into deployment rather than added at the end.

We believe Applied AI, not generative AI on its own, is where the next two to three years of competitive advantage will be decided. We're equally clear-eyed that most organisations aren't ready for it yet. That's why our engagements start with a readiness assessment, move to architecture, and only then to deployment. It's a slower first step, and it's the one that decides whether an agent ships once or keeps running unsupervised for years.

The real story is the scaling gap

The headline adoption numbers can create a misleading picture of progress. Despite widespread AI use, McKinsey found that nearly two-thirds of organisations haven't yet begun scaling AI across the enterprise. Only around 6% qualify as AI high performers, organisations that attribute at least 5% of EBIT and significant business value to their use of AI.

That gap isn't primarily caused by access to technology. Most organisations can use the same foundation models, cloud services, and enterprise platforms. A five-person startup and a global enterprise may have access to broadly similar model capabilities. What separates the organisations creating value is everything surrounding the model: a clear business objective, an operating model that supports AI at scale, reliable and governed data, and a process for measuring outcomes after deployment.

This is the pattern we see repeatedly across deployments. A pilot performs well; the initial demonstration generates enthusiasm, and then the solution meets real organisational conditions: inconsistent data, legacy integrations, security reviews, unclear ownership. Technology usually still works. What stalls is everything around it: the data foundation, the production infrastructure, and the identity and security model, none of which were designed to carry something that acts on decisions rather than just displaying them on a dashboard. Closing that gap is less about the model and more about integrating it properly into the cloud, data, and security environment an organisation already runs, as certified Google Cloud and AWS partners; that's the layer where most of our engineering time goes.

Agentic AI raises both the opportunity and the risk

If generative AI defined the last stage of adoption, agentic AI is defining the next one. Gartner predicts that by the end of 2026, up to 40% of enterprise applications will include integrated, task-specific AI agents, compared with less than 5% in 2025. These agents go beyond generating information. They can interpret objectives, plan multiple steps, interact with systems, and execute parts of a workflow.

That creates real opportunity for areas like IT operations, customer service, cybersecurity, and business-process automation. It also raises the stakes. An assistant that produces an inaccurate answer creates one kind of problem. An agent that acts on an inaccurate conclusion, changes a system, or triggers a business process creates another.

Gartner also predicts that more than 40% of agentic AI projects will be cancelled by the end of 2027, largely because of escalating costs, unclear business value, or inadequate risk controls. These two forecasts aren't contradictory. Rapid adoption and high failure rates can happen at the same time, often inside the same organisations. The lesson isn't to avoid agentic AI. It's to build it with production accountability from the start, so it can be monitored, audited, and rolled back rather than just demonstrated once and trusted to keep working.

None of this comes down to having a better model. Models are increasingly a commodity; the difference in outcomes comes from the architecture, governance, and operating discipline built around them. That's what closes the distance between a validated use case and a system an organisation can run, and it's the part of the work that rarely shows up in a product demo.

From adoption to operational readiness

For organisations that have already experimented with AI, the next question shouldn't be "what else can AI do?" It should be "what needs to be true about our data, operating model, and governance for this to run reliably at scale?" That question is harder than picking up a model or running another pilot, and it's the one that determines whether AI becomes a durable business capability or stays a collection of disconnected experiments.

Adoption is no longer the hard part. Getting the environment underneath it ready to carry the weight is.

"Every client conversation we have now starts the same way: they've already experimented with AI, and they want to know why it isn't paying off yet. The data backs up what we're seeing on the ground: McKinsey's 6% of AI high performers and Gartner's cancellation forecast for agentic projects are two sides of the same problem. Adoption was never going to be the hard part. We built our practice around the harder, less visible work of making AI safe enough to run without someone watching it every hour. That's the conversation worth having in 2026."

Greta Mieliauskaitė, Account Executive at TeraSky Europe

Sources:

• McKinsey & Company, "The State of AI in 2025: Agents, Innovation, and Transformation," November 2025.

• Gartner, "Gartner Predicts 40% of Enterprise Apps Will Feature Task-Specific AI Agents by 2026," August 2025.

• Gartner, "Gartner Predicts Over 40% of Agentic AI Projects Will Be Cancelled by End of 2027," June 2025.