Beyond AI anxiety: rethink, don't sprinkle

AI headlines are everywhere.

The pressure has moved from the lab to the boardroom, and it is changing how decisions get made. My inbox fills daily with vendors promising the next transformative AI tool. The instinct, under that much pressure, is to buy something fast, visible and reassuring.

I understand the instinct.

But I would argue for the opposite.

Don’t panic.

It remains very good advice, whether for intergalactic travel or enterprise AI.

Not because AI is overhyped. It is not.

But because panic is a terrible operating model.

Panic makes companies buy before they understand. It makes them sprinkle technology on top of broken workflows. It makes them confuse demos with capability, movement with progress, and adoption with advantage.

This is not a fear. It is already measured. MIT’s State of AI in Business 2025 study found that 95 percent of enterprise generative AI pilots delivered no measurable impact on the bottom line. The main cause was not the technology. It was organisational, what the researchers called a learning gap: the inability to integrate AI into actual workflows, structures and culture.

I spend time close to the frontier of this work, and the honest picture I see is that no one has fully solved enterprise grade agent management yet. Not the labs. Not the platforms. Google , for example, seems to be taking the right approach by building many of the foundations from the ground up: identity, access, governance, observability, cost control and accountability.

That is the part many vendor pitches skip. There is no shortcut to buy if the thing you need is organisational understanding.

Sothe answer to the anxiety is not more AI. It is less sprinkling and more rebuilding.

Competitive advantage will not come from layering agents on top of broken workflows. It will come from reimagining the workflow itself, with AI assumed from the start rather than bolted on at the end.

A few things matter more than the next demo.

First, domain understanding.

AI does not fix what you do not understand. If you cannot explain the problem clearly, an agent will simply automate your confusion at speed.

Second, internal technology depth.

This matters more than people admit. If you do not have enough AI and technology expertise inside the company, you will struggle to tell the difference between something genuinely useful and something that looks impressive in a vendor demo. You become easy to sell to.

Speaking to a few peer executives: everyone seems to have seen dozens of demos past 2-years, and maybe one or two were genuinely useful. The rest were wrappers or science projects

Third, governance that means something.

When an agent acts, someone owns the result. If everyone owns it, no one does. A human in the loop is only oversight if that human can actually evaluate what the agent did.

Fourth, cost clarity.

Token usage is becoming the new headcount bloat. It accrues quietly, across teams, with no one watching the total. Measure it now, or explain it later.

And underneath all of it, orchestration.

Identity, permissions, access, control, observability. No company has AI under control until acting, deciding and being accountable are wired together rather than assumed.

None of this is glamorous. All of it compounds.

Here is the part I find most interesting, and why I think some of the current anxiety is misplaced.

Most of what we are arguing about today still lives in language. Agents that read, write, summarise, reason and generate. That is powerful. But language is still only one slice of intelligence.

For those of us in physical industries, supply chains, products, stores, manufacturing, materials, movement, the real unlock is still ahead.

It arrives when AI understands space, not just text. When it grasps geometry, motion, touch, constraints and the three dimensional world the business actually operates in.

That is the frontier worth preparing for.

And the companies that will be ready are not the ones sprinkling fastest today. They are the ones rebuilding their understanding from first principles now, so there is something solid for the next wave to land on.

So the question I would ask is not:

“Which AI tool should we buy this quarter?”

It is:

“Which part of how we work would we rebuild from scratch if we assumed AI from day one?”