What 25 Years of Operations Leadership Taught Me About AI Adoption

AI adoption is the defining leadership challenge of this decade — and most organizations are failing at it badly. There are two types of people in operations right now: those who see new technology for its raw potential to solve problems, and those who see it as pure instability, something to be tolerated until it fails so they can go back to the old way. Most people fall somewhere in the middle. Like the early days of the .com era, we are in the volatile, high-stakes phase of adoption. The choices we make today will determine where our organizations stand in ten years, or if they exist at all.

I’ve spent twenty-five years in operations leading teams through broken processes, legacy migrations, and the kind of change management that makes people’s eyes glaze over at vendor demos. I focus on the pillars that actually drive multi-year change: the operational history of machine learning, agentic workflows, Personal Knowledge Management, Personal Context Management, and operational governance. I don’t care about product hype. I care about outcomes and stable solutions. My goal is to help leaders build predictable processes and refine them incrementally, changing workflows only when there is an undeniably good operational reason to move.

Here is what we need to avoid, and we have all seen it happen. A process bottleneck arises, you request vendor proposals, and the demo is incredibly slick. The PowerPoint slides are full of projected efficiency percentages. Everyone in the room nods at the right moments. Then the pilot runs for six months, produces a massive report nobody reads, and quietly dies on a shared drive.

I’ve watched that exact sequence play out more times than I can count. Different companies, different technologies, same result. The pilot worked fine. It was the organization that failed.

AI adoption isn’t a technology problem. It’s an operations problem. If you’ve spent decades leading teams and fixing broken processes, you are already carrying the exact tools you need.

The Numbers Are Sobering, But Not for the Reason You Think

The data on AI adoption is sobering. 95% of AI pilots never reach production. A separate RAND analysis of over 2,400 AI initiatives revealed that 80% never left the testing stage.

That isn’t a tech gap. The model doesn’t care if your approval workflow requires seven corporate signatures because of a mistake back in 2008. The model works fine. It’s the organization that’s broken.

The root causes aren’t technical; they are structural: poor data foundations, undefined ownership, zero deployment hypotheses, and change management bolted on as an afterthought. Every single one of those is a classic operations problem that an experienced leader has solved a hundred times before.

According to McKinsey’s State of AI, while 88% of organizations are utilizing AI in some capacity, barely a third have scaled it enterprise-wide. The gap between a successful pilot and daily operations isn’t a model selection issue. It’s an integration issue. That is the last mile, and operations leaders own the last mile.

What Twenty-Five Years of Operations Leadership Actually Builds

I came up through operations, managing fulfillment centers, leading teams through legacy system migrations nobody wanted, and watching what happens when you automate a broken process. Spoiler: you just get a faster broken process.

Eliyahu Goldratt’s Theory of Constraints taught me that local optimization kills systems. Making every department run at maximum isolated speed does not increase total throughput. The constraint runs the room. When I walk into an AI adoption strategy meeting, I’m asking the questions tech vendors avoid. Where is the bottleneck? What happens upstream when this specific piece accelerates? Is there even a real problem here, or are we just buying toys?

A technology gap is a resource problem with a resource solution. A leadership gap requires changing how people actually work in daily operations. And that always takes longer.

An experienced operator doesn’t just look at the tool; they see the entire system. They know exactly where a sudden burst of speed will break the workflow. But leadership gaps require changing how people actually work in daily operations. And that always takes longer.

When I Started Building My Own AI Stack

I stopped waiting for enterprise tech to reach frontline operators and started building my own local systems. I stood up a local LLM environment, a vector database with hybrid search, and what I call a Council of Rivals: eleven specialized AI profiles that each bring a different lens to an operational problem.

Underneath all of that, I run a personal knowledge system in Obsidian. It’s a 2,600-note vault structured so human intuition and AI agents can navigate the exact same data. I didn’t do this because I was a tech person at the time. My passion for what this technology could do drove me to figure it out, a journey that eventually led me to pursue my Master’s in Information Systems Management.

As I experimented, I realized that prompting is just another form of standard operating procedure. You provide the context, you set the parameters, and you build a system. That’s when I stumbled into Personal Context Management.

Here’s the truth: a vault with a thousand isolated notes is just a library no one visits. But a vault with connected, structured, intentionally enriched notes? That’s a thinking partner that gets smarter over time. The PKM community built these tools for basic note-taking, but what they are only starting to realize is that those well-structured notes, with consistent frontmatter and linked concepts, are accidentally optimized for AI consumption. You’ve been building a retrieval corpus without knowing it.

The practitioners pulling ahead aren’t the ones with the most expensive tech budgets. They’re the ones who built their knowledge architecture first, then layered the agents on top.

Every structured note I write feeds my RAG pipeline, sharpens my agent’s understanding of how I think, and compounds the quality of the execution.

The Discipline We Already Own

I served as a Fleet Marine Force Corpsman with Kilo Company, 3rd Battalion, 5th Marines. The Marine Corps didn’t teach me to fear chaos; it taught me to thrive on it, using frameworks like the OODA Loop to impose order faster than the environment could change. Twenty-five years of supply chain leadership reinforced that exact same lesson. You build the system, you work the system, and you improve the system. You seize the initiative instead of waiting for perfect conditions.

That’s exactly the discipline AI adoption requires right now. We don’t need aimless experimentation for curiosity’s sake. We need a deployment with a clear hypothesis. No more sprawling, endless pilots. What works is a constrained proof with a real owner and a defined path straight to production.

The AI boom isn’t a clean break from the past. It’s the next logical step for systems thinkers. For anyone who has spent twenty years managing processes, hunting down bottlenecks, and engineering workflows, this technology should feel completely familiar. It isn’t a revolution designed to rewrite your playbook. It’s a high-velocity tool built for the discipline we already own.