It was almost midnight and I was adjusting a shade of orange.
Not building anything. Not fixing anything a single reader would notice. Just nudging a hex value two digits one way, then back. On a site I'd already shipped. The front door to a practice that tells airlines to stop fussing over the front door.
I caught it eventually. The orange was the part I could see, so the orange got my time. The part underneath—the part that actually decides whether any of this works—had been waiting weeks for its turn. And I was doing the exact thing I get paid to talk airlines out of.
Built for the Platform
The problem is budgets reward what demos well. A redesign is easy for the board to see. A new data platform comes with a trip to Vegas or Orlando. You can point at them. The services and data underneath them are invisible until our investments fail to produce outcomes. So data waits.
You've seen the diagram. Website, tags, daily export, storage, a few stops along an architect's preferred pattern, then finally a report. Seven hops to move a click into a number. It crosses three cloud vendors on the way, because marketing bought one tool, data bought another and IT mandated a method—each a reasonable call on its own. But then the report's wrong, and nobody knows why. We blame the tools again and start over.
Take the services layer. A dev shop stands up an event-based architecture, wires it into a tool like Tibco. Now everyone agrees we have microservices under our site. But look at where those services were born—web design sessions and user stories, every service written to feed a page. What we get isn't a services layer for the business. It's middleware for a website.
Those tags are data in motion. Follow them downstream and they come to rest—the data layer. I've built bronze-silver-gold layers that earned their keep. That's not the problem. The trouble starts when the layers serve the platform's preferred pattern instead of the way the business actually runs. The warehouse holds every booking we've ever taken, but ask why last Tuesday softened and the honest answer is a three-day pull.
Built right, that question is self-serve. Shape the data around what the business asks—load factor, fare mix, why Tuesday softened—and the business answers itself, an analyst or an agent pulls the same number and nobody files a ticket to find out. "We have a medallion architecture" was never the win. "Anyone can answer their own questions from it" is.
Both these problems came out of the same room—one scoping the work with no one in it who'd ever filed a fare, settled a ticket or checked in a passenger with two bags and a lap child. You can't model the data around a question you've never had to answer, and you can't atomize an operation you've never run. That's a domain gap, not a tooling gap—and it's the whole job now, because leaders are judged on outcomes, not architectures.
Atomize How the Business Runs
The tools work. The build shipped. What's missing is the question that should have come first—how does the business actually run, broken into pieces any system can call and all of us can measure.
Take the things the business does—price a fare, hold a seat, recognize a customer, lift a coupon, settle a ticket—and expose each one as a service the business owns, sitting on data it can access and understand.
That's the foundation.
Build it once. Let anything consume it. The website is just one consumer. The mobile app is another. The next ones won't be a person at all. And when that happens, our websites become the phone book the year Yahoo! went mainstream.
The website is the easy part. Digital experiences. Data platforms. They’re not the hard part. Your data foundations are. In motion and at rest.
Build them. Own them.
What this means Monday morning
If you're a commercial analyst: Stop waiting in line behind the technology roadmap. You can't will the foundation into existence this quarter. But you can own one slice of it now. Take one thing your team needs—answer a revenue question, recognize a customer, hold a booking—and build the smallest working version on data and a service you control. Don't ask for the capability. Show it. Demonstrated capability moves an org that a roadmap request never will.
If you lead the data team: Stop grading the data layer by whether the medallion stack exists. Grade it by whether the business can answer its own question off it. Take one question your stakeholders keep asking by hand and make the data behind it clean, defined and owned—so the answer comes out the same every time and an agent could be trusted to pull it. Bronze-silver-gold isn't the win. A business that stops asking you for the number is.
If you're the executive funding the next move: Before the next redesign or platform, ask two questions. Can the business manage what we're building? Could an agent consume it anywhere? If the answer is no, you're buying surface—and your commercial team spends another quarter waiting in line behind it.
What's in the feed
I spent Wednesday at the AWS Summit NYC. Javits Center, every booth in the place selling an agent. Here are four moments with the same lesson:
Southwest CIO Lauren Butts, on Winter Storm Elliott: the systems did exactly what they were built to do. They just weren't built for everything hitting all at once. The system didn't fail. The design did. Built for a spec that no longer represented the business.
AWS's live AI demo pointed a desktop agent at a warehouse—Quick sampled Snowflake schemas, wrote its own SQL, built a dashboard live. Slick because the data underneath was clean. Aim it at a lake nobody curated and it writes confident SQL against noise. Quick is cool, but an agent doesn't fix the data layer. It inherits it.
Stripe showed what a coding agent services layer looks like: one central MCP server, ~1,600 tools, 1 million calls a day, feeding human engineers and agents off the same primitives. Built once, exposed once, consumed by anything—responsible for a platform processing $2 trillion a year.
The line of the day, from AWS's Chad Papa: telemetry without context is noise, telemetry with context is signal. True of every layer here. Whether in people or data—context, not intelligence, is value.
What I'm cooking this week
Not cooking. Eating. My son's birthday party was a throng of just-turned-teenagers. I catered the party by walking to the bodega and buying every empanada in the case.
The same place sells toilet paper and quarts of engine oil. They're not trying to look like anything—they just make one thing better than anyone in the neighborhood and have things people need, when they need them.
If airlines were in charge of the bodega, we'd have given it a redesign and lost the block. The look is the last thing that matters and the first thing we fund. Show me the lie.
Below the wing
Last week I said you can't reorganize your way to capability. This week, the same thing from the other side: you can't redesign your way to it either.
Above the wing is what the passenger sees. Below the wing is what makes the flight happen, and nobody photographs the ramp. The redesign, the new platform, the chatbot bolted onto your service—all above the wing. Most of the business is below it. Let’s fund it like we mean it.
—Ben