The AI revolution isn’t coming. It’s already here. According to the State of Application Strategy 2025 report, 96% of enterprises are deploying AI models. We’re well past chatbot trials and productivity co-pilots. These models are now driving workflow automation, security response and performance optimization. But while AI ambitions are on a tear, operations are still dragging their heels and, in many cases, blocking the path entirely.
Ops Debt: AI’s Silent Bottleneck
At least 60% of organizations are still locked into manual workflows, legacy ticketing systems and siloed tooling. That’s not just inconvenient, it’s operational debt. And it’s quietly throttling enterprise AI not because the models don’t work, but because the environment they rely on hasn’t evolved past 2010.
Think of it like this: AI is the rocket. Ops is the launchpad. Right now, most orgs are trying to launch next-gen tech off a moss-covered slab held together by duct tape and service tickets.
The tension shows up clearly in outcomes. Early analysis of our latest, AI-focused research reveals that “increased productivity” and “improved operational efficiency” are the two most commonly realized benefits of AI, cited by 71% and 64% of respondents, respectively. That’s great. But if AI is delivering efficiency everywhere except IT operations, it raises a question: what’s broken?
Spoiler: it’s the process.
Even organizations that want to automate can’t, because their processes are as fragmented as their environments. Nearly a quarter cite ticketing system integration alone as a top blocker. And when we asked what’s actually consuming automation teams’ time, the top answers weren’t “fine-tuning LLMs” or “building agents”. No, they were managing vendor APIs and cranking out glue code. Ops teams are drowning in integration ductwork instead of shipping solutions.
This disconnect feeds directly into the talent crunch. It’s not just a lack of AI skillsets, it’s a lack of headroom. The people who could be scaling AI are too busy troubleshooting pipelines and waiting for approval to run scripts. Only 45% of orgs report any automated execution of app delivery or security tasks, despite nearly all saying they’re comfortable letting AI take the wheel on at least one operational function.
The punchline? AI is ready to help. It’s the environment that’s not ready to receive it.
Automation Alone Won’t Save You
Enter observability-driven automation and AIOps. This year, for the first time, automation – not alerting – is the top use case for operational telemetry. Two-thirds of respondents are already automating based on telemetry data, and more are close behind. That’s a sign of progress, but automation alone won’t save you from fragmentation.
Nearly 80% of enterprises are running hybrid. They’re juggling a median of four public cloud vendors. That kind of sprawl is a nightmare for consistency unless you build a programmable, cloud-agnostic foundation. The most forward-thinking orgs are doing exactly that: cutting API bloat, consolidating toolsets and deploying orchestration platforms that treat apps, infrastructure and AI models as one operational plane.
But let’s be clear: Programmability isn’t a “nice to have.” It’s a survival trait.
The State of Application Strategy 2025 found that the #1 use case for programmability is mitigating zero-day threats. And with attackers increasingly using generative AI to find and exploit vulnerabilities faster, zero-days are about to become the new normal. In that reality, the ability to inject policy or rewrite traffic in real time, without waiting for human approval, doesn’t just matter. It’s table stakes.
AI Maturity Requires Infrastructure Maturity
Nowhere is the maturity gap more obvious than in the climb from GenAI to Agentic AI. Our focused research shows nearly half of organizations (47%) are running GenAI in full production. But when it comes to AI Agents? That drops to 9%. And Agentic AI? Just 5%.
It’s not a lack of interest holding them back. 50% say they’re still in early stages with AI Agents and 37% say the same for Agentic AI. The problem isn’t imagination. It’s infrastructure.
Agentic AI needs more than intelligence; it also needs autonomy. That means real-time telemetry, dynamic policy execution, secure API access and programmable platforms that can handle OODA loops (Observe, Orient, Decide, Act) without tripping over legacy systems. Most enterprises? Not there yet.
To scale AI, enterprises must invest in automated pipelines, cloud-agnostic architecture and programmable security. This isn’t about talent or budgets alone, it’s about aligning operations with ambition. Infrastructure isn’t just AI’s foundation; it’s the gateway to its future. Until it evolves, AI’s full potential remains out of reach.
