TL;DR
- Enterprise AI has moved decisively past the pilot phase, with worker access up 50% in 2025 and production deployments accelerating fast.
- Only 1 in 5 companies has a mature governance model for agentic AI — even as autonomous systems take on real financial, operational, and customer-facing decisions.
- The talent gap remains the single biggest barrier: a shortage of qualified AI experts is stalling adoption across every industry.
- The enterprises pulling ahead aren't just buying tools — they're rebuilding the foundations of data, process, and skilled implementation that make AI sustainable.
Welcome to the Scaling Era. Ready or Not.
There's a moment in every technology revolution when the question shifts from "Should we try this?" to "Why aren't we already doing this at scale?" For enterprise AI, that moment arrived sometime in the last twelve months — and it didn't send a calendar invite.
Deloitte's State of AI in the Enterprise report makes the inflection point impossible to ignore: worker access to AI tools rose 50% in 2025, and the share of companies running 40% or more of their AI experiments in full production is on track to double within six months. The pilot era — that comfortable, low-stakes period of sandboxed experiments and PowerPoint decks titled "Our AI Journey" — is over. The scaling era has begun. And it is already exposing a gap wide enough to drive a data center through.
The Ambition-Execution Chasm
Here's the uncomfortable math: 74% of enterprises say they want AI to grow revenue. Only 20% say it actually has. That's not a technology problem. That's an execution problem wearing a technology costume.
The bottleneck isn't ambition, budget, or even access to models. According to NVIDIA's survey of more than 3,200 organizations, the single biggest barrier to AI adoption — across every industry surveyed — is the lack of qualified AI experts. Not compute. Not cost. People. The irony is almost poetic: companies are racing to deploy systems designed to augment human intelligence, and they're being slowed down by a shortage of human intelligence to deploy them properly.
Meanwhile, a stubborn 37% of companies are still running AI at a surface level — bolting a chatbot onto an existing portal, auto-completing a few emails, and calling it a transformation. Underneath? The workflows, data pipelines, and organizational structures are exactly as they were in 2022. The paint is AI-colored. The house is the same.
Agentic AI: The Governance Time Bomb
If surface-level deployment is the slow lane, agentic AI is the Autobahn — and many enterprises are driving it without functioning brakes.
Agentic AI refers to autonomous systems capable of independently executing complex decisions: flagging and resolving financial discrepancies, rerouting supply chains in real time, handling multi-step customer service interactions without a human in the loop. These aren't experimental curiosities anymore. They are live, in production, making consequential calls.
And yet, only 1 in 5 companies has a mature governance model to oversee them. That means 80% of enterprises are, to varying degrees, running autonomous decision-making systems under frameworks that weren't designed for autonomous decision-making systems. That's a bit like installing autopilot in a car and then not updating the traffic laws.
"The question is no longer whether AI can make decisions autonomously. The question is whether your organization is structured to take responsibility for those decisions."
Governance here isn't red tape for its own sake. It's the difference between AI that scales trust and AI that scales risk. Mature governance models define accountability chains, establish audit trails, set override protocols, and ensure that when an agentic system makes a bad call — and eventually, one will — there's a human and a process ready to catch it.
Where Generative AI Is Actually Transforming IT
While the governance debate plays out in boardrooms, Generative AI is already doing real, unglamorous, high-value work inside IT departments. And it's worth understanding where, because this is where the gap between surface-level adopters and serious scalers becomes visible.
Code generation and development acceleration is one of the clearest wins. AI coding tools aren't just autocompleting variable names — they're generating functional code snippets, surfacing bugs before they hit review, and auto-generating documentation that developers have historically written approximately never. Development cycles that used to take weeks are compressing, and the quality floor is rising.
Incident response and IT operations are seeing similar transformation. Generative AI can ingest system logs at a scale no human team can match, identify anomalies in real time, correlate patterns across infrastructure layers, and surface actionable recommendations before a slow query becomes a catastrophic outage. Predictive maintenance — once a concept reserved for manufacturing floors — is becoming standard in enterprise IT.
Knowledge management, long one of IT's most persistent headaches, is another frontier. In large organizations, institutional knowledge is scattered across wikis no one updates, ticketing systems no one searches, and the heads of people who just gave two weeks' notice. Generative AI can synthesize, surface, and contextualize that information on demand — meaning the answer to "how do we handle this edge case in the legacy billing system" doesn't have to live exclusively in Dave's brain anymore. Sorry, Dave.
Intelligent service desks are perhaps the most visible change for end users. AI-powered virtual assistants can resolve common technical issues instantly, triage complex ones intelligently, and learn from interaction histories to improve over time. IT teams that once spent 60% of their capacity on repetitive Tier-1 tickets are reclaiming that time for work that actually requires human judgment.
The Traits of Enterprises That Are Actually Winning
Across every sector, the enterprises that are successfully navigating the scaling era share a recognizable set of traits — and none of them are "they bought the most expensive AI platform."
1. They treat data as infrastructure, not an afterthought.
AI is only as good as the data it operates on. Organizations that invested in data quality, lineage, and governance before AI became urgent are now moving faster because they're not simultaneously trying to clean up ten years of messy data lakes while training models on top of them.
2. They redesign processes before automating them.
Automating a broken process doesn't fix it — it just breaks it faster, at scale. The enterprises pulling ahead mapped their workflows first, identified where AI could genuinely add value, and redesigned around that — rather than wrapping AI around the existing mess and hoping for the best.
3. They invest in AI literacy across the organization, not just in a dedicated team.
The talent gap is real, but the most effective response isn't just hiring more AI specialists. It's building enough fluency throughout the organization that managers, analysts, and functional leaders can engage meaningfully with AI systems, identify problems, and advocate for better implementations. You don't need every employee to build models. You need every employee to understand what the model is actually doing.
4. They take governance seriously before something goes wrong.
Mature AI governance isn't reactive. By the time you need it urgently, you're already dealing with consequences. The leading enterprises are building governance frameworks now — defining accountability, establishing oversight mechanisms, and stress-testing their agentic systems — not because regulators are forcing them to, but because they understand the downside of not doing so.
The Bridge Problem — and Who Solves It
Here's the underlying tension that the scaling era is surfacing: most enterprises know where they want to go, and most AI tools can theoretically get them there. The gap is the bridge between those two things — the implementation expertise, organizational change management, data architecture, and ongoing optimization that turn a promising AI deployment into a durable competitive advantage.
This is not a small gap. And it is not a gap that closes itself.
For IT services companies, consultancies, and managed service providers, this inflection point is essentially a clarion call. The enterprises that can't build that bridge internally — which is most of them — need partners who can. Not vendors who sell licenses and walk away, but implementation partners who understand both the technology and the messy, human, process-laden reality of deploying it inside complex organizations.
The scaling era rewards depth. Superficial AI adoption is table stakes; anyone with a credit card and an API key can have it. What's genuinely scarce — and increasingly valuable — is the expertise to make AI work in the real conditions of real enterprises: legacy systems, regulatory constraints, change-resistant cultures, and all.
The Bottom Line
Enterprise AI in 2026 is not a question of whether — it's a question of how well. The organizations that treat the scaling era as a procurement exercise will end up with expensive tools and modest results. The ones that treat it as a fundamental transformation — of their data, their processes, their people, and their governance — will build something that actually compounds over time.
The pilot era was about proving AI could work. The scaling era is about proving your organization can. That's a much harder test. And most enterprises, if they're being honest, haven't fully started studying for it yet.
The gap between ambition and execution won't close on its own. But at least now, we know exactly where it is.
Published in Stream · Dispatch #440 · July 1, 2026 · 8 min read.
Reply to paolo@mont3.ch - every email gets a human answer within 24h.