Foundations · Field notes
When an AI project stalls, everyone blames the model. Usually the technology isn’t the problem at all — the foundations are.
We had a conversation this week with a company trying to guide their business and people through the world of AI. They weren’t short on ideas, or appetite. Everyone wants to “use AI” right now — that’s not the challenge. But as we talked, a familiar pattern emerged. Whenever a project stalled or slipped into confusion, the blame was always placed on the AI itself:
“The model isn’t accurate enough.” “The tools aren’t mature yet.” “We’re waiting for the next release.”
The more we dug into it, the more obvious it became that the technology wasn’t the problem at all. AI wasn’t failing them — their foundations were.
In our experience, and with most companies, AI adoption starts with good intentions and ends with unplanned sprawl. Individual teams try different tools in isolation — marketing use one tool, operations use another, customer service find their own solution — and before long the organisation has half a dozen AI tools and LLMs in play with no visibility of how they relate to each other. Nothing connects, nobody owns it, and the risks creep in quietly.
Then there’s the overconfidence in quick-fix approaches. Uploading a few documents into a RAG pipeline or dropping a chatbot on the website might look impressive in a demo, but that doesn’t mean it works across real processes. The moment you introduce sensitive data, multi-step workflows, responsibility for outcomes, or regulatory scrutiny, the limitations show.
It’s not because the AI can’t handle the task. It’s because the organisation hasn’t put any structure around how AI is meant to operate. When we walked through our Conductor platform and approach — the orchestration, the audit layer, the ability to route tasks properly, the integration with existing systems, the visibility over what’s happening — the conversation shifted.
They didn’t need “a better model.” They needed a framework that made AI dependable — something that handled the plumbing, the governance, and the operational reality that sits around the intelligence itself.
This is exactly the point where so many AI projects fail. Not at the prototype stage, where everything looks neat and isolated, but at the moment a business tries to move from a demo to something that affects real people, real customers, or real decisions.
If there’s one thing that came out of that conversation, it’s this: AI isn’t letting companies down. Companies are letting themselves down by trying to scale AI on top of weak foundations.
The organisations making real progress aren’t the ones experimenting the most or buying the flashiest tools. They’re the ones doing the unglamorous work first: agreeing ownership, setting guardrails, connecting systems properly, defining when AI leads and when a human needs to step in, and making sure everything is observable and auditable. Once those pieces are in place, AI becomes far easier to deploy, far safer to trust, and far more valuable across the business.
Don’t start by asking what AI can do. Start by asking whether your organisation is actually ready to run it. Get the foundations right, and the AI will do the rest.
Not sure whether your foundations are ready to scale AI? That’s exactly what a discovery call is for.
