It's Not the AI. It's Us.

A conversation with Jose Luis Loren triggered something that had been sitting right in front of me for months.

We were talking about AI implementation at work — what makes it stick, what makes it fail — and somewhere in that conversation, something clicked. The problem isn’t the technology. It’s never been the technology.

A Prosci study of over 1,000 professionals found that 63% of organizations cite human factors as the primary challenge in AI implementation. Not infrastructure. Not cost. Not capability. People.

I nodded when I first saw that stat. But I didn’t quite see the full picture.

In high-traffic analytics environments, the pattern is consistent: when new infrastructure gets introduced — a new tool, an MCP layer, a server-side setup, a new data source — the people you’d expect to push back rarely do. The analysts get it immediately. The friction comes from the organisational layers around them. The meetings before the meetings. The approvals. The “what even is this” conversations with stakeholders who don’t yet have a frame of reference for what they’re being asked to evaluate.

The analysts got it immediately. Because they live in this space. They already understand why the new thing is better.

So when we say “the human factor”, we need to be more precise about which humans.

A few months ago I wrote about how my role has never fit neatly into traditional org charts. I’ve been doing something like it for over ten years, across industries from travel to fintech to iGaming, and in every company the shape of the friction was similar: I wasn’t quite a developer, wasn’t quite a marketer, wasn’t quite a data engineer. I lived in the intersections.

For a long time, that was the awkward part of the job.

Now I think it’s the competitive advantage.

Before AI, the value of a technical profile was the technical skill itself. If you could write the query, build the pipeline, configure the tag — you were necessary. The skill was the moat.

AI is eroding that moat. Not completely. Not for everyone. But to a meaningful degree.

A junior marketer can now ask a reasonably configured AI to write SQL and get a working query. A product manager can use an MCP-connected tool to pull their own GA4 data without knowing what a dimension is. The floor for technical execution has dropped. The bar for contributing technically is lower than it has ever been.

What AI is still not good at is generating context.

RAG changes what AI can access — give it the right documents, the right data sources, the right knowledge base, and it can retrieve relevant context on demand. That’s real and it matters.

But someone has to build that knowledge base. Someone has to decide what goes in, what’s outdated, and what’s trustworthy. And the most valuable organizational context — the kind that actually drives decisions — often doesn’t exist in any document at all.

Why does that metric look different in this dashboard than in that one? Why does a traffic spike in this market mean something completely different than the same spike somewhere else? Why is the question the stakeholder is asking not actually the question they need answered?

That knowledge lives in people. Specifically, in people who have spent years paying attention across the whole system. RAG can retrieve what was written down. It can’t retrieve what was never written — because the person who knew it just knew.

That’s what I mean by transversal thinking. Not a technical skill. Not a soft skill. A capacity for abstraction across scopes — one that operates on a different logic than deep technical expertise. It’s not better. It’s different. And it’s the thing that’s hard to automate.

The developer optimises within the system. The data engineer builds the pipes. The IT architect designs the infrastructure. All of them essential. But the person who has to evaluate the output across all of those systems, translate it into business language, and surface the insight that actually drives a decision — that person is doing something AI can assist but not replace.

Because the AI doesn’t know that the spike in organic traffic this week is probably related to a campaign that launched in a different team last Tuesday. The AI doesn’t know the CFO is asking about ROAS because there’s a budget review next week and the real question is about channel allocation. The AI doesn’t know that the “increase in new users” in this particular market is almost certainly bot traffic.

You know. Because you’ve been paying attention across the whole system.

I’m not making an argument against specialists. Deep technical expertise is more valuable than ever — the people who can build the AI layer, evaluate its outputs, and fix it when it fails are not going anywhere. But they were never going to be automated out in the first place.

What I’m saying is that the profiles who were always somewhere between specialist and generalist — who were always evaluated on their ability to synthesize across contexts, not just execute within one — those profiles now have structural leverage they didn’t have before.

The AI closed the technical gap. It hasn’t closed the context gap.

And context, built over years of transversal work, is the thing that’s hardest to replicate.

This isn’t a comfortable argument to make. It can sound like reassurance for people like me — like I’m telling myself I’m safe. I’m not sure “safe” is the right frame. What I’m saying is narrower: the skill that made us awkward fits in traditional org structures is exactly the skill that makes us valuable in AI-native ones.

The problem was never AI.

The problem was that organisations weren’t built for this kind of thinking.

Maybe now they’ll have to be.

These are my personal opinions. Not my employer’s. I’ve been wrong before — feel free to tell me I’m wrong again.