74% aspire to revenue with AI. Only 20% achieve it. The Deloitte report explains the gap.

Every quarter a new report on AI in the enterprise appears. Most say the same thing: "adoption is growing". Deloitte's "State of AI in the Enterprise 2026" says something more uncomfortable: the distance between what companies aspire to achieve with AI and what they actually achieve is widening. 74% aspire to increase revenue with AI. Only 20% achieve it. And the reason isn't technical.

This analysis walks through the report's most relevant data and connects each finding with what we observe in real implementations. The numbers tell exactly the same story we've been seeing in development teams: the technology works. The organizational system isn't ready to take advantage of it.

The numbers that define 2026: accelerated adoption, uneven results

Before getting into the analysis, it's worth sizing the moment. Deloitte's report gathers data from hundreds of organizations globally, from startups to large corporations. The headlines paint a picture of mass adoption, but with nuances that matter:

The headlines are positive. Access to AI tools among workers grew 50% in 2025. Companies with more than 40% of projects in production will double in the next six months. 42% of organizations believe their strategy is "highly prepared".

But look at the other numbers. Only 1 in 5 has mature governance for autonomous AI. Only 34% is reimagining its business, not just optimizing processes. And there's one figure that's the key to the whole report: 74% of companies aspire to increase revenue with AI, but only 20% are achieving it.

That 54-point difference isn't a technical gap. It's an organizational gap.

54 points of difference: the biggest gap isn't technological

This is the report's central figure and it's worth pausing on.

In productivity and efficiency, the gap is manageable: 66% report real gains. Same with improved decisions (53%) and cost reduction (40%). These are the benefits of "doing the same thing, faster".

But when we move to business outcomes -- generating more revenue, improving products, innovating -- the gap explodes. Only 20% achieve more revenue. Only 20% improve products or services. Only 38% improve the relationship with customers.

The pattern is clear: AI is serving to optimize operations, but not to transform the business. And that isn't a problem with the technology. It's a problem with how organizations are framing adoption.

Most companies use AI to make what they already did more efficient. Only a minority is reimagining what they do.

This connects with a data point we already analyzed in another context: 70% of AI success in development teams is organizational, not technical. Deloitte's report validates it with enterprise-scale data.

66% optimize. Only 34% transform. The difference is structural.

Of all the report's data, this is perhaps the one that best explains the gap: only 34% of organizations is "truly reimagining its business" with AI. The rest -- two thirds -- use it mainly for operational efficiency.

This doesn't mean operational efficiency is bad. The 66% reporting productivity gains is getting real value. The 53% improving decisions is making use of data it wasn't processing before. Those results are legitimate.

The problem is that staying in efficiency has a ceiling. And that ceiling arrives fast.

Deloitte's report points to a key reason for this distribution: companies report significantly lower preparation in infrastructure, data management, risk frameworks and talent capabilities than in strategy.

In other words: at the PowerPoint level, the strategy exists. At the execution level, the foundations aren't in place.

42% believe their strategy is "highly prepared". But when you drill down to the operational detail -- the infrastructure to support AI in production, data with the necessary quality, teams with the right skills, risk frameworks to manage autonomous agents -- preparation plummets.

There's a gap between strategic preparation and operational preparation. And it's in that gap that most organizations get trapped: they know what they want to do with AI, but they don't have the foundations to do it.

Agentic AI: adoption is skyrocketing, governance isn't keeping up

The report notes that the use of agentic AI is set for abrupt growth over the next two years. And at the same time, only 1 in 5 companies has mature governance for autonomous agents.

This is a worrying imbalance.

Agentic AI isn't a more powerful version of a chatbot. They're systems that make decisions, execute actions and operate with varying degrees of autonomy. Without adequate governance, this means code deployed without sufficient oversight, decisions made without clear traceability and risks nobody has mapped.

We already analyzed in detail the situation of AI agents in companies for 2026. Deloitte's data confirms the same pattern from another angle: the obstacle isn't deploying agents. It's governing them.

For development teams, this translates into very concrete questions:

• If an agent generates code and deploys it, who is responsible for quality?
• If an agent makes architectural decisions, is there traceability of why?
• If an agent operates autonomously in a CI/CD pipeline, what limits does it have?
• If something fails in production and an agent generated it, how is it diagnosed?

Without a work system that answers these questions, agentic AI goes from being a competitive advantage to being an unmanaged operational risk.

Related reading: In our Agentic AI analysis we detail what it means for an agent to be truly autonomous and how it differs from conventional automation. And in code quality in the AI coding era, we address the specific review criteria for agent-generated code.

The skills gap is real. But the solution isn't one-off training.

Deloitte's report identifies the AI skills gap as the main barrier to adoption. So far, nothing new. What's relevant is how companies are responding.

According to Deloitte, organizations prioritize education and training over the redesign of roles and workflows. That is: the dominant response to the skills gap is to run courses, not to change how teams work.

And this explains why, despite access to AI tools growing 50% in 2025, the impact on the workforce remains low. More people have access. Fewer people know what to do with that access within their daily workflow.

This is a pattern we already addressed in depth: why teams don't adopt the AI tools you buy them. Access isn't the problem. Integration into the workflow is.

The skills gap is a symptom, not a cause. The cause is that organizations are trying to fit new capabilities into old work systems. It's like giving someone a racing car and expecting them to go faster around the city without changing the routes, the traffic lights or the traffic rules.

The skills that are missing aren't technical in the traditional sense. It isn't that people don't know how to write a prompt. It's that the roles, processes and quality criteria adapted for those skills to be applied systematically don't exist.

From diagnosis to system: how SDD and Centers of Excellence address each gap

Deloitte's report describes three main deficits: operational preparation below strategic preparation, insufficient governance for agentic AI, and a skills gap treated with training instead of redesign. Each of these deficits has a direct correlate in how we work at onext.

1. The strategy-execution gap is closed with a work system, not a roadmap.

42% of companies say their strategy is prepared. But when you drill down to the operational detail -- data, infrastructure, talent -- preparation collapses. This happens because the strategy stays at the level of "what we want to do" without landing on "how we're going to work differently".

Spec-Driven Development (SDD) exists to close that gap. Instead of generic strategies, SDD defines concrete specifications: the project "constitution" sets rules the AI must respect, the spec templates standardize how features are defined for agents, and the prompt playbook turns good intentions into repeatable practices. It isn't a strategy. It's an operating system for teams that work with AI.

2. Agentic AI governance starts with traceability and clear limits.

Only 1 in 5 companies has mature governance. For the rest, the risk isn't abstract: it's code in production without adequate review, decisions without traceability and diffuse responsibilities.

The answer isn't to add a layer of bureaucracy. It's to integrate governance into the workflow. In SDD, every line of generated code is traceable to a specification. The code review flow has specific criteria for agent output. And context engineering ensures agents operate with the right context, not generic or incomplete context. Governance not as bureaucracy, but as system architecture.

3. The skills gap is solved with Centers of Excellence, not courses.

Deloitte notes that companies prioritize education over role redesign. An AI Center of Excellence does the opposite: it first redesigns the work system and then upskills the team within the new system. Training isn't the goal. It's a consequence of a workflow that naturally demands new skills.

It's no coincidence that the three gaps converge on the same thing. The common denominator is organizational: a work system designed for AI is missing. The tools are there. The models are good enough. What's missing is the intermediate layer between "we have an AI strategy" and "our team works differently thanks to AI".

The report confirms the thesis. Execution depends on the system.

Deloitte's "State of AI in the Enterprise 2026" doesn't say anything the teams already in production with AI didn't know intuitively. But it quantifies it. And the numbers are hard to ignore.

54 points of difference between aspiration and real revenue. Two thirds of companies stuck in operational efficiency. Only 1 in 5 with governance for autonomous agents. A skills gap treated with courses instead of structural change.

The good news: none of this is an unsolvable problem. It's a system problem. And systems get redesigned.

The technology is already here. The models are enough. The question that remains is organizational: is your work system designed so that AI generates real impact, or just so that people have access to tools?