AI is making the gap worse, not better

For 30 years I have watched the same thing happen in good organisations. Leadership sets a clear direction. Smart, committed people go to work. And somewhere between the strategy offsite and the actual outcome, momentum quietly dies.

It is rarely a talent problem. It is almost always a dialogue problem — three gaps that nobody owns:

No dialogue between intent and action. The "what" is announced; the "how" is assumed.
No legitimate channel for disagreement. The people closest to the work can see the plan won't survive contact with reality, and have nowhere safe to say so.
No feedback loop. By the time the gap is visible in the numbers, the quarter is already gone.

I spent five years building a software-engineering platform organisation inside a regulated pharma company, and a good part of the last year helping a team rebuild a product with AI agents. In both, the gap was the real enemy — not the technology.

So here is my uncomfortable claim

For most organisations, the current wave of AI adoption is making that gap worse, not better.

Not because the tools are bad. They are remarkable. But because we are layering automation on top of a broken operating model and calling it transformation. When you do that, three things happen:

You accelerate the wrong thing. A process that was quietly misaligned now produces misaligned output faster, and with more confidence.
Judgment quietly drains out of the system. Each individual hand-off to the machine is reasonable. The aggregate is an organisation where fewer and fewer people can say why a decision was made — only that the tool suggested it.
You create a governance debt you can't see. Agent use spreads ad hoc, faster than anyone can govern it — right as the EU AI Act, the Cyber Resilience Act, NIS2 and GDPR turn "we'll sort it later" into real deadlines and real penalties.

The seductive pitch is that AI does the heavy lifting so your people are freed up. That is true, and it is also the least interesting thing about it. Framing AI as toil relief is how you end up automating a mess.

The principle that changes it

The fix is not more AI, or less. It is a rule about who decides:

AI proposes. Humans decide. And every decision traces back to its source.

This isn't a compromise to keep the lawyers happy. It is how high-performing teams have always worked, long before the technology. Tell people what needs to be achieved and why. Let them — and now their tools — work out the how. Keep a legitimate channel for them to push back, because disagreement is signal, not insubordination. And keep the thread of why intact, so that when someone asks "how did we get here?", there is an answer.

I believe this strongly enough that I have built it into software. In the compliance system I work on, the AI writes the query and phrases the answer — but it can only tell you what a live query actually returned, every claim cited back to the regulation that backs it, right down to the article. Then, before it answers, it tries to falsify itself. The AI is the interface to the truth, not the source of it — and the irreversible actions still need a human signature. In the strategy tooling, the AI facilitates and quality-checks, but it never decides direction, and nothing moves to the next layer until a human has approved the last. Same principle, twice: humans think, AI lifts.

What this means on Monday

If you lead an engineering or compliance organisation adopting AI, you do not need a bigger model. You need three things most rollouts skip:

A clear statement of intent people can actually act on — and permission to disagree with it.
A deliberate line between what the AI proposes and what a human decides — made explicit, not left to each individual's judgment under deadline.
Traceability by design, so adoption is defensible the day a regulator, an auditor, or your own board asks.

Do that, and AI stops widening the gap between intent and outcomes and starts closing it. Skip it, and you will simply arrive at the wrong place faster — with a very confident audit trail of how you got there, if you are lucky enough to have one at all.

I'm Tete Mensa-Annan. I spent 21 years at Microsoft and five building the software engineering platform at Novo Nordisk; now I help engineering and compliance leaders adopt AI without losing control of it. If the gap between intent and outcomes sounds familiar, I'm always up for a genuine conversation about where it lives in your organisation — not a pitch.