The engineering-ification of everything

Marketing used to be an art.

In the 1960s, the job was: come up with a clever campaign, run it, and hope. The best marketers were the ones with the best instincts—people who could feel what would resonate. Measurement was crude. You ran an ad in a newspaper and tried to guess what it did to sales.

Today, marketing is an engineering discipline. A/B testing, attribution models, funnel analytics, conversion optimization, programmatic ad buying. The instinct is still there in some roles, but the center of gravity has shifted. The question isn't "does this feel right?" It's "what does the data say?"

Sales followed the same arc. It used to be pure relationship craft—the good salesperson just knew how to read a room. Now it's CRM pipelines, lead scoring, cadence optimization, and conversion analytics. The best salespeople still have instinct, but they operate inside systems that measure everything.

This isn't a story about marketing and sales. It's a pattern that keeps showing up, field after field, decade after decade.

Weather forecasting

This might be the cleanest example historically.

Experienced meteorologists used to read cloud formations, barometric pressure, and wind patterns with intuition built over decades of practice. A good forecaster had a feel for what was coming. The job was closer to craft than computation.

Then computational fluid dynamics, satellite data, and ensemble modeling arrived. Today, a weather forecast is an engineering output. The craft layer is almost entirely gone. And accuracy has improved dramatically—not because forecasters got better at reading clouds, but because the system got instrumented.

Agriculture

One of the oldest crafts on earth. For centuries, farming ran on generational knowledge: when to plant, how to rotate crops, what the soil needed. An experienced farmer could read the land.

Now it's precision agriculture. GPS-guided equipment, soil sensors, satellite imagery, yield modeling, drone surveys. The experienced farmer still exists, but the best operations are run by people who combine that experience with data from systems that see things no human eye can.

Sports

The Moneyball moment made this visible to everyone: a baseball team started winning by ignoring scouts' instincts and following statistical models instead.

But it's gone far beyond that. Modern sports teams use wearable sensors, GPS tracking, biomechanical analysis, sleep and recovery data, and game-film analytics processed by machine learning. Coaching is increasingly an engineering discipline with a craft layer on top, not the other way around. The "eye for talent" hasn't disappeared, but it now operates inside a system that measures everything the eye can't.

Movies and series

Hollywood used to run on mogul instinct. A studio head greenlit films because they "knew" what audiences wanted. Star power was the main variable. TV networks picked up shows based on pilot episodes and executive taste. Test screenings existed, but the decision-making was largely gut.

Now it's franchise formulas optimized for global box office, test screenings with real-time audience biometrics, and streaming platforms greenlighting series based on viewing pattern data. Netflix doesn't have a programming executive with great taste deciding what gets a second season. It has engagement data on 200 million subscribers telling it exactly where people pause, skip, rewatch, and drop off. The shift from "a producer who feels it" to "a system that predicts it" happened in about two decades—and it hit series even harder than film, because series generate continuous engagement data that a two-hour movie never could.

Politics

Campaign strategy used to be party machines, stump speeches, and experienced operatives who knew their districts. The best campaign managers had instinct for what messages would land and where.

Obama 2008 was the inflection point. Micro-targeting, voter data modeling, A/B-tested messaging, real-time sentiment tracking. Every serious campaign since has been a data operation with a political layer on top. The experienced operative still matters, but they now work inside systems that can test ten messages before breakfast and tell you which one moves likely voters in swing counties.

Dating

This one is almost absurd in how stark the shift is.

Finding a partner used to be entirely social—your network, your neighborhood, chance encounters at work or school or through friends. The "system" was just life, and the skill was social intuition.

What apps like Tinder actually did was make a market visible. You're not being matched by a clever algorithm—you're browsing a marketplace of people who are also browsing, and mutual interest is the filter. The entire dynamic that used to be implicit and slow (do they like me? should I approach? what's the signal?) got collapsed into a swipe. Supply and demand, surfaced and measurable. The platforms then optimize on top of that: who to show you, in what order, based on what gets engagement. Whether the engineering has made dating better is a separate question. But a domain that was pure social craft is now substantially mediated by a system that instruments and optimizes human preference at scale.

The mechanism

The pattern across all of these is the same.

Instrumentation comes first. Optimization follows.

In every case, the shift started when someone figured out how to measure the thing that used to be felt. Cloud patterns became fluid dynamics equations. Talent scouting became performance metrics. Campaign instinct became voter models. A producer's taste became viewing data.

Once you can measure it, engineering thinking moves in—not because engineers are taking over, but because measurement changes what "good" means. It moves from "someone experienced thinks so" to "the data shows it works." And once that transition happens, it doesn't reverse. The craft layer shrinks, the engineering layer grows, and the field compounds faster because solutions become transferable and improvable instead of trapped inside individual heads.

What resists (and why)

Some fields have held out longer than others. And even within a single field, you can see the shift happen unevenly—depending on one variable: how cheap is verification?

Education is a good case study because it's not a monolith. Parts of it have already been engineered. Software engineering education shifted early—bootcamps, MOOCs, structured online courses—because verification is trivial: did the code run? Did it produce the right output? You can check that automatically, at scale, instantly. The same is true for much of math and the hard sciences, where the answer is either right or wrong and a machine can tell.

Language learning has been harder. Duolingo made the most serious attempt at engineering-ifying it, and got far on vocabulary and grammar drills where verification is still relatively cheap. But the harder parts—conversation, nuance, fluency—resisted, because verifying whether someone is actually communicating well in a new language used to require a human listener.

The pattern is consistent: engineering-ification advances in education exactly as fast as verification gets cheaper. Where verification is easy, the field has already shifted. Where verification requires human judgment, it stalls.

AI compresses the timeline

Every field I listed above took decades to shift. The instrumentation had to be built, the data infrastructure had to mature, and the analytical tools had to catch up to the complexity of the domain.

AI changes the speed of this process—specifically because it makes verification cheap in domains where it wasn't before.

Language learning is the immediate example. Verifying whether someone can hold a conversation in French used to require a French speaker sitting across from them. Now an AI can do it, in real time, at negligible cost. The verification bottleneck that kept language learning craft-heavy just disappeared.

The same logic applies everywhere verification was the bottleneck. The fields that haven't been engineered yet aren't exempt from this pattern. They were just waiting for cheap verification. And that's arriving now.