We are our own biggest barrier to AI transformation

Most organizations aren’t ready for the breakthrough potential of AI.

AI agents can now draft thousands of marketing messages, test them across simulated audiences, and surface the top few for human review. Satellite-based models can track migration through ambient light, fuel consumption, and shipping data. Code-generating tools write, test, and debug in a fraction of the time it takes a human.

I think we will see the first gains of the modern workplace in areas of AI-augmented or AI-replaced creativity. This is AI basically doing the things we already do, but better and faster.

But the deeper transformation—how AI reshapes organizations—will stall unless we address a core issue: we’re simply not organized enough to take advantage of it.

We need to invest in knowledge management before we can see AI transformation

“Knowledge management” refers to everything from where and how we organize and store key information, to the corresponding processes and structures for accessing and learning from our experiences. This is how teams document their work, communicate about it, and–most importantly–learn from their collective experiences. Naturally, thinking about AI transformation often starts with trying to teach AI the same way we teach each other: sharing experiences, stories of important company history, anecdotes that represent key elements of company culture or ethos, and macro-level data. AI doesn't learn like humans, and this approach consistently fails.

I’ve seen countless AI pilots stall out before they start—because the knowledge systems just aren’t ready. Files are missing. Data units don’t match. Teams reinvent the wheel on tasks ripe for automation (NIH progress reports, anyone?). Sometimes this is for understandable and justifiable reasons: funders have different reporting structures, different regulatory environments require different information, or different teams–especially those working across languages–have different standards and practices.

At one organization, there was no central repository for current contracts—despite hundreds of millions in annual revenue (and it was impossible to get awards from the past 5, 10, or 20 years). A government health system in Latin America had entirely separate data systems for workplace-related health issues, injuries, cancers, and all other hospital encounters. AI can't navigate fragmented systems like these without enormous manual effort.

One approach might be to just let an AI have access to all our knowledge, in its current disorganized state, but giving AI access to disorganized knowledge doesn’t work. In one pilot, we trained a chatbot on hundreds of project folders. AI couldn’t distinguish “final_workplan.docx” from “/archive/old-notes-9-months-ago.docx.” Teaching AI to navigate one team’s quirks is hard enough—scaling that effort is unsustainable.

Many AI experts underestimate the messiness of organizational life. Their experience in software engineering—where development follows structured steps—doesn’t translate cleanly to the messy management of people, projects, or institutions.

Barriers to preparing our systems for AI transformation

So what makes AI preparation so hard? Three things stand out:

Barrier 1: Legacy systems

Systems exist for a reason. From simple details like which desk to route an internal help-ticket, to the complicated custom implementations of an EHR, we often have spent huge sums of money and invested significant training efforts to strategically structure our knowledge and processes to serve our current needs. But with AI, the landscape has changed, and we’ll need to take a hard look at when we might need to change our practices versus trying to fit AI on top of existing norms.

Barrier 2: Current systems aren’t broken

Most systems work. And that’s the problem. The AI wave feels like electrification or the internet: a seismic shift that makes current tools look outdated—but still functional. Refactoring live systems is risky, and business doesn't pause for folder reorganization. Any effort to change fully functioning systems to prepare for AI transformation will, by definition, be disruptive. Clients don’t have much patience for delays because of re-organizing your folder structure. Deadlines don’t shift because of code refactors. Taxes are still due, regardless of our financial system overhaul. 

Additionally, we have often built specific internal tools on top of the existing structures - that one graph the CEO likes to see, a custom PowerBI dashboard for HR leaders, automated emails or slack messages from Jira. It is hard to change our systems and processes to be more AI-friendly when the current systems are functioning well and meeting our needs.

Barrier 3: Humans don’t like change

Ultimately, knowledge management is human behavior. It reflects how teams think, prioritize, and share their work. I once helped grow an org from 150 to 300 staff in two years. We moved from “make a spreadsheet” to “there’s a system for that”—but adoption lagged. People found ways around the system, often with good reason.

Why use an impersonal helpdesk when your favorite IT person sits next to you? Why submit a form when you control your budget? These are human choices, and they don't change just because a new tool exists.

We cannot underestimate the human side of AI transformation, starting first with knowledge management. Asking team members to change where they put something or how they do something is not just asking them to use a new tool, it is asking them to think differently about the very work they do. This is unsettling. This is hard. This is slow.

But this is not impossible; and it is necessary.

Strategies to prepare for AI transformation

Start small. Pick a contained function or team and get that data ecosystem in shape. Maybe don’t overhaul your entire patient billing system—but let a pilot group use an AI chatbot to answer common billing questions.

In healthcare, watch for transformation in large, integrated verticals—where data is already standardized and workflows are high-volume. These systems are primed for AI acceleration.

This is why I am watching things like stryker acquiring care.ai, Kaiser rolling out Abridge, or Blue Cross adding Tempus.AI to their network.

But small companies, group practices, or local health centers can pick a single function and see what AI-augmentation might bring. Ideally this is something where data and information is already standardized, contained, and clean. The goal is to demonstrate value, which then motivates future investments in AI transformation projects.

Some argue that “starting small” is too slow for this moment. I disagree. People—and institutions—don’t change overnight. Real AI transformation will require thinking fundamentally differently about how we organize our knowledge, share experiences, and do our daily work. This cannot happen overnight, and certainly not with top-down mandates.

Leaders should pair urgency with empathy. Change is uncomfortable. Support it with clear goals, useful tools, and a deep respect for the people doing the work.

Concluding thoughts

AI won’t realize its full promise until we overhaul how we manage knowledge. Our systems were built for humans—messy, creative, inconsistent. AI needs clarity, consistency, and structure.

Yes, this work is hard. It’s expensive. It challenges existing systems and behaviors. But it’s also necessary—and transformative.

Start with what’s standardized. Prove value with targeted roll-outs that demonstrate value. And above all, lead with empathy.

The AI revolution is already here. The question is whether we’re ready to meet it.