The instant-coffee problem with AI

A few years ago, I worked on a large capital project where we were trying to bring a then-emerging technology into a complex operational business. Everybody in the market was launching similar projects then. The deck looked impressive. The renders looked impressive. The vision pitched well in the boardroom.

We failed.

Not because the technology didn't work. We failed because we did not understand where this technology actually fit in the business model. We were in love with the technology. We forgot to look inside the business. We could not tie the economic outcomes to the transformation we were leading. So the project quietly died, and now it sits on a list of failed experiments.

The same thing is happening with agentic AI today, at a much larger scale. And the failure mode is identical.

Most enterprises bringing in AI today are not acting on a need. They are acting on what the competition is doing. The moment 20 tweets a day mention AI agents, the CEO suddenly wants to do AI. The mandate passes down. Managers scramble to demonstrate progress. Engineers get pulled into projects without a clear problem statement. Vendors get into the building.

Nobody pauses to think about what the actual problem is.

People think in pieces. Bring in AI first. Solve the other problems as they come up. Then the next problem. Then the next one after that. Nobody sits down and creates an image, in absolute detail, of what the transformation looks like end to end — what the before state is, what the after state should be, and how the business actually gets from one to the other.

That picture in detail is what transformation actually means. Without it, you are not transforming anything. You are buying technology.

And technology, alone, has never been the bottleneck.

An organisation is not just buildings and people. Every organisation has a meta-brain — the sum total of all the interactions, all the processes, all the formal workflows, and crucially, all the informal ones. How communication actually happens. How approvals actually flow. How exceptions actually get handled. How things actually get done on the ground, regardless of what the policies and HR manuals say.

This meta-brain is the real shape of the business. Most of it isn't documented anywhere. It lives in the implicit knowledge of senior operators, in the workarounds people use when the official system doesn't fit reality.

When you bring AI into an organisation without understanding this meta-brain, the technology doesn't fit. It stands there like a stump. It doesn't serve any function. And because the organisation doesn't know how to assimilate it, the people don't adopt it. The project becomes another expensive experiment in the failure pile.

This is not a technology problem. It is a context problem.

After watching enterprises get this wrong consistently — and after being part of one of those failures myself — I have come to believe there are three things that matter more than any technology choice.

First — always think context before technology. Tech is there to solve the business, not the other way around. Before any AI agent is built, before any pilot is scoped, before any vendor is engaged, somebody senior in the organisation has to sit down and answer the unglamorous questions. How is work actually done here? What are the formal processes, and what are the informal ones? Where are the bottlenecks people complain about every quarter? Where is the operating model already straining? Which workflow, if it changed, would the CEO notice within ninety days?

This is harder work than buying technology. It is also the only work that determines whether the technology will produce operating change or just produce pilots.

Second — governance is not an afterthought. It is the core of the transformation. If governance is added at the end — after the agent is built, after the pilot is running, after something has already gone wrong — it is never going to fully contain the system you built. It will always be patchwork. Where does the human sit in the loop? What are the escalation thresholds? How is performance measured against a baseline that existed before the agent? Where does the audit trail live? Who is accountable when the agent is wrong? These questions get answered at day zero, or they get answered in court.

Third — keep people at the centre. Technology is here to solve for people, not to replace people. If you do not bring the people on board — if you do not understand what value they add today, how their work is going to change, what new roles they will move into — the transformation will fail no matter how good the model is. People are not an afterthought. They are the assimilation surface. Without them, nothing sticks.

A transformation done this way takes longer than the demo-driven version. It produces a different kind of artifact.

Not a slide deck describing success. A measured before-and-after of a real workflow. A documented architecture with human checkpoints designed in from day zero. A governance model that survives an audit. A runbook for the operating team. A failure log, kept honestly, with the lessons that came from it. An expansion plan for adjacent workflows. And the capability inside the client team to run the next deployment without the consultant in the room.

That is operating-model change. The transformations that produce it look very different from the transformations that don't.

This work requires senior decision-makers who already understand that AI matters and who will not tolerate vague strategy exercises or vendor theater. It requires the willingness to baseline first — to measure honestly before deploying anything. It requires accepting that the most credible operating systems are built deliberately, not assembled quickly. And it requires the patience to do unglamorous work before the impressive work.

Most enterprises will not do this. They will continue to accumulate pilots. They will continue to license copilots without redesigning what the workforce does. They will continue to demo well, ship rarely, and never alter how work is done.

The enterprises that do this work — context first, governance baked in from day zero, people at the centre — will have something the others don't.

A different operating model. One that compounds.

Caden Tech is a practice for enterprises moving past AI pilots toward measured operating-model transformation. Based in UAE, serving clients globally.

To discuss whether your organisation is ready for an agentic operating-model engagement, book a discovery call or write to [email protected].