When AI goes down, what goes with it?

Amazon's AI systems went down earlier this year, and businesses across sectors discovered how deeply they had embedded AI into operations they assumed were solid. Customer service queues stalled. Logistics decisions slowed. Some organisations could not process orders at all.

The outage lasted hours. The operational conversations it opened will take much longer to close.

The dependency you may not have mapped

Most businesses that have adopted AI tools over the past three years have done so incrementally. A tool here for customer queries. An integration there for inventory. A model running in the background to flag anomalies. Each adoption made sense individually. Few organisations stepped back to map what they had built in aggregate.

When the Amazon outage hit, the picture became clear. Organisations with AI deeply embedded in their operations found themselves more fragile than they had understood themselves to be.

This is not an argument against AI adoption. It is an argument for understanding what you are actually building.

Scaling what is already there

One of the more uncomfortable findings from organisations that have studied AI implementation failures is this: AI does not transform operational culture. It amplifies it.

A manufacturing business we studied in The Stillness Dividend had invested significantly in AI-driven process optimisation. The tools were sophisticated. The implementation was technically competent. Costs rose anyway. Throughput fell.

The reason was not the AI. The reason was that the organisation had real misalignments in how decisions were made, how priorities were set, and how teams communicated. The AI accelerated those processes, and amplified those misalignments along with everything else.

The finding is blunt: "AI doesn't solve incoherence. It scales it."

The Amazon outage exposed a version of this problem at infrastructure level. The more common version plays out quietly inside organisations every week, in the gap between what AI produces and what the business actually needed.

The cost of rework

When AI output is misaligned with organisational intent, the work has to be done again. In The Stillness Dividend, we call this the Rework Tax: the compounding cost of effort that produces output that then has to be corrected, redone, or discarded.

Research with a London architecture firm found that 38% of project work was being redone. That translated to a seven-figure annual cost. The reason was not incompetence. The speed of delivery had pulled ahead of the quality of decisions upstream. AI tends to accelerate delivery. If the decisions upstream are unclear, it accelerates the production of work that will need to be undone.

What AI reliability actually requires

Technical reliability - uptime, API stability, model consistency - matters. But it is only one layer of the question.

The deeper question is whether your organisation has the coherence to use AI well when the systems are working. That means:

• Clear decision rights: who owns which outputs, and who is accountable for acting on them

• Defined review processes: where human judgement must be applied before AI output is acted on

• Honest mapping of dependencies: what would stop if a specific AI tool went down for 24 hours

The Amazon outage is worth treating as a drill. Run the scenario for your own operations. The organisations that find the exercise uncomfortable are often the ones with the most to address.

The governance question is now operational

AI governance used to feel like a compliance or ethics conversation. It is increasingly an operational one. Organisations with clear protocols for how AI is used, reviewed, and overridden are more resilient to outages, to model drift, and to the quieter failures that do not make headlines.

The starting point is an honest assessment of how coherent your current AI deployment actually is. Not whether the tools work, but whether the organisation around them is ready to use them well.

Is your leadership team ready to deploy AI effectively? Read our guide to leadership team AI readiness and what it actually takes.

Stillness Partners works with leadership teams and boards to build the organisational coherence that makes AI implementation pay. If the question of AI reliability is live in your business, speak to us.