I’ve watched three Fortune 100 companies bolt AI onto their existing processes in the last eighteen months and wonder why the ROI didn’t materialize. The tools worked. The models performed. The demos were impressive. And the operating cost barely moved.
The problem wasn’t the AI. It was the process the AI was asked to serve.
The process is the problem
When a manufacturer adds AI to sort defect images on a production line, something real has happened. A human step — slow, inconsistent, expensive — has been replaced by a faster, cheaper, more reliable one. That’s value. But the production line itself is unchanged. The bottlenecks before and after the AI step are still human-scale. The throughput is still governed by a process architecture designed around what people can do.
This is what most “AI transformation” looks like today. It’s process improvement — sometimes dramatic process improvement — but it is not transformation. The process was designed for a world without AI, and adding AI to it is like putting a jet engine on a bicycle. You’ll go faster. You might also realize the bicycle was the wrong vehicle.
Three postures
I’ve come to think about an organization’s relationship to AI in terms of three postures. They are not phases — you don’t have to progress through them in order — but they describe materially different ceilings on what AI can do for you.
Posture one: AI as laborer. You tell it how to do something. Sort these images. Extract these fields. Route these tickets using these rules. The process stays the same; a human step is swapped for a cheaper one. The ceiling is the ceiling of the existing process.
Posture two: AI as junior employee. You tell it what to achieve. Flag anomalous transactions. Summarize these contracts. Draft a response to this RFP. AI has latitude to choose its method, and the results are often meaningfully better than posture one. This is where most enterprises are today, and it’s legitimate. But the process architecture — designed for humans, with human handoffs and human bottlenecks — is still the frame AI operates inside.
Posture three: AI as intent partner. You express your intent — your goals and objectives — and AI achieves them. Not “flag anomalous transactions” but “ensure this portfolio’s risk stays within these bounds given these market conditions.” Not “sort defect images” but “maintain product quality at this standard across this production line.” At this posture, AI can redesign the process itself. The operating model no longer assumes human bottlenecks as permanent constraints. Complexity drops by orders of magnitude.
The objection
The most common pushback I hear: “That sounds like AGI. We’re not there yet.”
Fair. But the gap between posture two and posture three is not a technology gap — it’s a design gap. The models available today can operate at posture three if the operating model is designed for it. The reason most organizations are stuck at posture two isn’t that the AI isn’t capable. It’s that the process was designed for humans, and nobody has redesigned it for intent-driven AI.
When we stood up a full data governance program and MDM solution in four months — from scratch, with zero license costs — the AI wasn’t magic. The models were commercially available. What was different was the architecture. We didn’t ask “how do we use AI to speed up the existing MDM process?” We asked “what does a data governance program look like when AI is a first-class participant from day one?” The answer looked nothing like the 18-month vendor-driven model everyone else was running.
Four months. Zero license cost. The method is the difference, not the model.
The test
Here’s a question you can ask in your next meeting about an AI initiative: “Are we telling AI what to do, or telling AI what we want?”
If the answer is the former — if every AI interaction starts with an instruction rather than an objective — you’re at posture two at best. That’s fine as a starting point. It is not fine as the destination.
The organizations that will pull away over the next three years are the ones that redesign their operating models around intent, not the ones that add AI to their existing instructions. The difference is not incremental. It is structural.
What to do Monday
Pick one process in your organization where AI is already deployed at posture one or two. Ask: what would this look like if we expressed intent instead of instructions? Not “use AI to speed up the approval workflow” but “ensure qualified applications are approved within these risk bounds.”
You don’t need new technology for this exercise. You need new architecture. And the architecture starts with the question.