A 300-Year-Old Brewery and the Future of Your Company

In December 2024, Jeff Bezos told a story at the NYT DealBook Summit that I keep coming back to. He visited a 300-year-old brewery in Luxembourg that had a museum inside. In that museum was a hundred-year-old electric generator. The brewery had built its own power station to electrify its operations. There was no grid — if you wanted electricity, you generated it yourself.

Bezos’s analogy was direct: “That’s what computation was like — everyone had their own data center. That won’t last. You’ll buy compute off the grid. That’s AWS.” And now, the same pattern is repeating with AI.

Most B2B companies I work with are treating AI exactly the way that brewery treated electricity a century ago: as a one-off installation for a specific use. A chatbot here, a text generator there. Not as the grid that should be powering everything.

AI is a horizontal layer, not a vertical product

The most quoted line from that interview was this: “Modern AI is a horizontal enabling layer. It can be used to improve everything. It will be in everything. This is most like electricity.”

The distinction between horizontal and vertical matters. A vertical product solves one specific problem: whether it’s a chatbot for inquiries, a ticket classifier, or an image generator. Each one lives in its own silo. A horizontal layer, on the other hand, redefines how every process it touches actually works.

In our experience at IQ Source, the companies that extract the most value from AI aren’t the ones that bought the best tool — they’re the ones that rethought how work flows between departments. The difference is in the starting point. A company asking “what’s the best chatbot?” is buying a product. A company asking “where is effort duplicated across sales, operations, and support?” is designing infrastructure.

When you see AI as a product, you buy a license, assign it to a team, and measure its isolated ROI. When you see it as infrastructure, you identify every friction point in operations and assess where a layer of language processing, classification, or analysis reduces manual work. The result isn’t a new tool — it’s a company that operates differently.

That’s exactly the approach behind our AI assessment: before recommending any technology, we map complete processes to understand where AI generates real impact and where it just creates noise.

Amazon is working on a thousand AI applications — not one

When Andrew Ross Sorkin asked Bezos what he was working on, the answer was revealing: “It’s 95 percent AI.” And then he added: “We’re literally working on a thousand applications internally.”

That’s not rhetorical exaggeration. Amazon applies AI to warehouse logistics, delivery route optimization, fraud detection, dynamic pricing, customer service, developer productivity, content generation, contract review, and hundreds more. There’s no “AI department” managing one big project — AI is woven through the entire operation.

Do you need Amazon’s budget to do something similar? No. You need their mindset.

The exercise is simpler than it sounds: take each department in your company — sales, finance, operations, HR, support — and ask one question per department: where do people spend the most time making decisions that follow a repeatable pattern? Qualifying leads, reviewing documents, answering frequent questions, generating reports, prioritizing tickets. Every one of those tasks is a candidate for an AI layer.

We wrote a department-by-department guide for exactly that exercise. It’s not about installing a language model at every desk — it’s about identifying where smart automation removes real friction.

The difference between Amazon and the average B2B company isn’t technological. It’s that Amazon didn’t treat AI as a project with an end date. They treated it as an operational capability applied to everything they do.

The isolated pilot mistake

The most common pattern we see is this: a company decides to try AI. They pick a safe department — usually support or marketing — and implement one specific tool. If it works, they celebrate. If it doesn’t, they conclude “AI isn’t for us.”

Both conclusions are wrong.

A successful pilot in one department tells you nothing about AI’s value as infrastructure. It tells you that particular tool solved that particular problem. A failed pilot doesn’t prove AI doesn’t work either — it proves that specific use case, with that configuration, didn’t deliver results.

The electricity analogy makes this tangible. Imagine a factory in 1905 that electrified a single machine and kept everything else running on steam. The electric machine probably outperformed its steam equivalent. But the factory didn’t transform. The real transformation of electricity in manufacturing didn’t arrive until factories were completely redesigned around individual electric motors: new floor plans, new workflows, possibilities that didn’t even exist with centralized steam power.

AI works the same way. A chatbot might reduce first-tier support tickets by 30%. Useful, yes. But the real impact shows up when ticket data automatically feeds product priorities, query patterns update your knowledge base, and sales teams get signals about which features customers request most. That’s not a chatbot — that’s infrastructure.

We’ve seen companies get disappointed with their AI pilot not because the technology failed, but because they scoped it so narrowly that it never had the chance to show its real value.

What this means in practice

Shifting from “AI as tool” to “AI as infrastructure” doesn’t require a massive budget. It requires a change in how you approach the problem. These are the four decisions we see making the difference:

Start with friction, not technology. Don’t begin with “where can we use AI?” Start with “where do people spend the most time on repetitive decisions?” Classify those points by department. The result is an opportunity map prioritized by operational impact, not by technological novelty. Our implementation guide details how to run that mapping step by step.

Think in capabilities, not products. Instead of buying a chatbot for support, a text generator for marketing, and a classifier for sales, identify cross-cutting capabilities: document processing, automatic classification, content generation, pattern analysis. Apply each capability to the departments that need it. Fewer vendors, broader coverage.

Leadership has to show up. Bezos dedicates 95% of his time to AI. Not every CEO needs to do the same, but if AI adoption is delegated to a committee without budget or authority to change processes, it won’t move forward. The signal has to come from the top — with priority, resources, and clear expectations.

Measure like infrastructure, not like a project. The question isn’t “did the chatbot save 10 hours a month?” The question is “how much operational friction was reduced across the company?” The right metric measures workflow, not individual tools. Cycle time, rework rate, cross-department response speed — those are the indicators that AI is working as a layer, not as an accessory.

The real risk is standing still

Another Bezos quote worth keeping: “It’s human nature to overestimate risk and underestimate opportunity.”

The risk of an AI pilot that doesn’t deliver is a few months of effort invested. The risk of treating AI as optional — something you’ll “look at next year” — is that your competitors aren’t waiting. And I don’t mean Silicon Valley companies. I mean your direct competition, the ones already using AI to speed up quote responses, improve lead qualification, and cut document processing from hours to minutes.

AI API costs dropped over 80% between 2024 and 2026, according to Andreessen Horowitz. The barrier is no longer budget. It’s the decision to stop treating AI as an experiment and start treating it as what Bezos describes: this generation’s electricity.

If your company is still running isolated pilots, we can help you see the full picture. Start with our AI maturity assessment to understand where you stand today, or let’s talk directly about a cross-department audit that covers your entire operation.