Don’t Get Distracted By Apps—the True Power Of Ai Is …
In the spring of 1893, a 24-year-old engineer named William Henry Merrill arrived in Chicago with an unglamorous assignment. The World’s Columbian Exposition was about to show off electric light on a scale nobody had seen, with more than 100,000 incandescent lamps. There was just one little problem — the new technology was not only new, it was dangerous, and fires kept breaking out unexpectedly.
Merrill realized that to become truly useful, the electricity system would have to be designed as, well, a system. The ing year he ran his first product test in a small lab above a Chicago fire patrol station. That lab became Underwriters Laboratories, and it was instrumental in creating system-wide standards. By 1897, insurers, engineers, and the electrical industry had agreed on the first National Electrical Code.
Neither the lab nor the code produced a single watt of power. But electricity could not have fundamentally transformed our lives without them.
I’ve been thinking about Merrill since a recent conversation with Leigh-Ann Russell, the chief technology officer and global head of engineering at BNY, America’s oldest bank. Russell is a mechanical engineer who spent 18 years at BP before moving into banking, and she brought some of the energy industry’s vocabulary with her. One word in particular has stayed with me: interlock.
The device that says no
An interlock stops a system from operating until the conditions for operating it safely have been met. Railways developed the idea so that signals and switches could not be set against each other. In 1922, with London electrifying and switchgear injuring and killing the people who worked on it, an engineer named James Harry Castell borrowed the concept from the railways. In his design, the key that opens an equipment cabinet stays trapped in the switch for as long as the power is on. Nobody has to remember the rule, because the hardware won’t allow anything else.
Shipping learned the same lesson at a terrible price. In March 1987 the ferry Herald of Free Enterprise left Zeebrugge with its bow doors open and capsized, killing 193 people. The doors could not be seen from the bridge and there was no indicator light, although captains had asked for one. Just over a year later, international regulators adopted a rule requiring door indicators on the bridge.
Russell’s point is that industries these do not wait until the plant is running to ask what happens when one component misbehaves. And they never assume that because every part works, the whole will.
“It’s clear to me that there is a generational opportunity taking place right now with AI and, other transformational technology before it, the early innings are critical. However, there will be many inflection points. Phase one was adoption and now we’re moving to adaption. To engineering a system that both compounds in value over time but also reinvents itself,” Russell said. “To me, this is the moment to be principled about what we build, who it benefits, but most critically how it builds trust.”
Edison’s real invention
What Thomas Edison switched on at Pearl Street in lower Manhattan on September 4, 1882, was a complete system: dynamos, 80,000 feet of underground conductors, switches, sockets, meters to bill customers, and safety fuses to keep a fault in one building from becoming a fire in the next. His team had to design most of it because none of it existed. The historian Thomas Hughes built a classic study, Networks of Power, around this observation: Edison’s achievement was the system.
Even with the system in place, the payoff was slow. The economist Paul David showed that by 1899, almost two decades after Pearl Street, electric motors supplied less than 5% of the mechanical drives in American factories. Owners who did electrify usually swapped the steam engine for one large motor and left the shafts and belts where they were. The productivity surge came in the 1920s, once factories were redesigned around a small motor on every machine.
Which brings us to where we are with AI. Most organizations are in the swap-the-engine phase with AI. Models are better, code is cheap, and a team can prototype in an afternoon what used to take a quarter. So one team builds an agent, another deploys a chatbot, a third pilots a workflow tool. Each is clever and locally useful. Together they add up to fragmented workflows, inconsistent controls, duplicated costs, new security exposure, and a technology estate that someone has to run. Gartner predicts that more than 40% of agentic AI projects will be canceled by the end of 2027, and the reasons it gives are escalating costs, unclear business value, and inadequate risk controls.
Four disciplines, wired together
Russell says that for years she has asked the same questions of any growth opportunity: Will customers love it? Can we build it? Will it create value? Does it survive in the broader ecosystem, or is it a point solution? With AI, she argues, an application can pass all four and still leave the organization more expensive, more fragile, and harder to govern. So she adds a fifth: Can we build and operate it safely as part of a larger system?
Her answer is to design governance, security, verification, and observability into the architecture from the start, and she organizes that work around four disciplines.
Intent. Start with a real problem. When building gets cheap, deciding what not to build becomes more valuable.
Integrity. Engineer trust into the system: clear boundaries, meaningful evaluation, visibility into behavior, and working mechanisms for a human to step in. I think of this as the AI equivalent of the fuse box.
Infrastructure. Which model is best matters less than the layers around it: data, integration, compute, controls, and monitoring. d, reusable foundations beat a collection of bespoke wins. This is the equivalent of a safe distribution system.
Impact. Ask what changed. Did customers get a better experience? Did productivity improve? Did risk go down? Did the benefits cover the full cost of integrating, operating, and governing the thing? This is the AI version of the meter. Edison had to invent that too.
What makes this an interlock and not a checklist is that the four depend on one another. Intent shapes the infrastructure. Infrastructure makes integrity practical. Integrity is what allows responsible deployment, and impact tells you whether the original intent was sound.
What leaders should take from this
Count systems, not demos. Building something visible has never been cheaper, which makes visible activity a poor measure of progress. The better question is whether each new tool makes a contribution to the whole, not just to an isolated problem.
Treat governance as engineering. Nobody regards a circuit breaker as useless bureaucracy. The electrical code is the reason you can plug an appliance into an outlet without thinking about it. d AI foundations should do the same job, so that each new use case is faster to deploy and cheaper to operate because the hard parts are already designed and instrumented.
Measure the whole. A use case that succeeds locally can impose costs elsewhere. Judge AI by what it does to the full operating model.
Leave room to rewire. The factories that won in the 1920s could rearrange their floors as demand changed. An AI architecture so tightly coupled that one change forces a rebuild of everything else will not age well.
We have spent the past few years asking what AI can do. Russell thinks the next questions are harder: What should we build, how should we engineer it, how do we operate it safely, and how will we know it was worth doing?
Nobody remembers which exhibitor had the brightest lamps in Chicago in 1893. The electrical code that came out of that decade is still being revised and enforced today. Merrill’s motto was “Know by test, and state the facts,” and it would serve an AI program well.
Sumber Artikel:
Fastcompany.com