Experts shed light on how model-driven engineering can help programmers solve problems and reduce software errors more effectively than traditional programming
There were six hours during the night of April 10, 2014, when the entire population of Washington State had no 911 service. Tweets: @timbray , @zelbinian , @jsomers , @tyronecoupland , and @adriennelaf Tweets: Tim Bray / @timbray : Hm, thinking I need five thousand words or so to express my thoughts on this. Reading non-coders' words about coding is... weird. http://twitter.com/... Dustin / @zelbinian : “Visual Studio [has] 55 million lines of code. And ... [the study found] >98% of it is completely irrelevant.” #ux https://www.theatlantic.com/ ... James Somers / @jsomers : Started working on this story about why software fails, and how to fix it, two years ago—and now it's finally here! https://www.theatlantic.com/ ... Tyrone Coupland / @tyronecoupland : “Software doesn't break...Software failures are failures of understanding, and of imagination.” @jsomers Atlantic https://www.theatlantic.com/ ... Adrienne LaFrance / @adriennelaf : “The people know *how* to code. The problem is *what* to code.” Phenomenal piece by the great @jsomers. https://www.theatlantic.com/ ...
Context & Ripple Effects
James Somers' Atlantic feature anchors its argument in a concrete failure: the six-hour statewide 911 outage that hit Washington State on April 10, 2014, when the entire population lost emergency service overnight. The expert analysis he surfaces holds that model-driven engineering — raising the artifact from hand-written lines to models — catches errors traditional programming lets through.
The piece lands inside an older argument the related coverage keeps returning to: the software industry builds excellent tools for itself while everyone else gets bad software (doctors, academics, and scientists stuck with poor tools), so any method promising fewer errors is really a question about who gets to build reliable systems.
First-order effects
- Programmers evaluating model-driven engineering gain a concrete selling point — error reduction — backed by the kind of high-stakes failure the Washington State 911 outage represents, shifting the debate from developer preference to system reliability.
- Teams maintaining large legacy codebases, of the scale the article cites in Visual Studio's 55 million lines, face pressure to justify why the overwhelming majority of that code is irrelevant to the behavior users actually experience.
Second-order effects
- Carmakers packing 100-150+ chip-based electronic control units into each vehicle (an industry increasingly dependent on software) become natural adopters of higher-abstraction methods, since error rates compound across that many embedded systems.
- If models rather than code become the primary artifact, tool vendors compete on modeling environments instead of editors and compilers, echoing the integration gaps already documented in cloud development tooling.
Third-order effects
- A development culture built on models and generated artifacts rather than line-by-line craft converges with what is happening to juniors leaning on Copilot and Claude (foundational coding knowledge going missing) — both point toward engineers supervising abstractions they did not write by hand.
- Safety-critical public infrastructure like 911 systems becomes the proving ground for whether abstraction-first methods deliver on reliability, with regulators and procurement bodies as the eventual arbiters if the pattern holds.
The trend: Software development is steadily migrating up the abstraction stack — from hand-written code to models and now AI-assisted generation — with error reduction as the recurring justification at each step.