Tom Enden shipped production code he didn't fully understand, written in a programming language he's not fluent in — and stood on stage at Wix Engineering Conference 2026 admitting it. A year earlier, he says, this would have been a fireable offense. Last week, it was just another Tuesday.

The Hallway Feeling Nobody Talks About

There's a quiet existential dread circulating through every major tech company right now: if AI writes all the code, what exactly are engineers for? Enden doesn't sugarcoat it — anyone claiming they're not scared is probably lying. But his point isn't that we're doomed or safe. It's that our job has fundamentally changed in three ways, and understanding those shifts separates engineers who'll thrive from those who become commodity labor.

Shift I: From Monologue to Dialogue

Writing code used to mean opening an editor and facing a blinking cursor — pure monologue. Now it's prompting, iterating, pushing back against the model. Enden draws a sharp distinction between issuing commands ("make it work," "ship it") and conducting actual Socratic dialogue ("why that approach over a simpler interface?", "what breaks if traffic is 10×?"). The first scenario makes you a customer dropping a car at a mechanic. The second makes you an engineer. This cognitive shift — from typing to conversing — requires presence, curiosity, and holding a coherent mental model while interrogating the output.

Shift II: Distributed Authorship, Undivided Ownership

Roland Barthes wrote in 1967 that any text is a tissue of quotations; there is no original, there never was. Enden applies this directly to modern AI-assisted development: authorship has always been distributed through design patterns absorbed from others, Stack Overflow answers, and dependencies written by strangers you've never met. The single author was always fiction — now we just can't sustain it anymore. But Vercel's litmus test cuts through the philosophy: "Would you be comfortable owning a production incident tied to this pull request?" That's not about explaining every line; it's about whether you'd take the 3am page. Relying on AI means outsourcing your thinking. Leveraging it means you own the outcome, hold the mental model, and stand behind it.

Shift III: Taste Is the Work

If how we write changed and who writes changed — what remains that's uniquely ours? Enden's answer is blunt: taste and judgment. Greg Brockman of OpenAI calls taste a "new core skill." Paul Graham frames it as knowing what to build when anyone can build anything — an LLM generates a thousand solutions, only you know which one fits your system, team, and moment. Robert C. Martin (Uncle Bob) puts it plainly: we're losing syntax, and good riddance. The less brain space consumed by semicolons and braces, the more room for harder problems.

The Next Abstraction Rung

Every previous leap — assembly to C, C to managed languages, managed languages to frameworks — triggered warnings that "this isn't real engineering." Each time, critics were wrong. But Enden acknowledges this step feels different: those transitions had gates (CS degrees, comfort with code). AI-assisted development doesn't. As Andrej Karpathy put it years ago, the hottest new programming language is English. Anyone can build software now — doctors, teachers, children. That means engineers are standing on the same rung as everyone who speaks English. The difference? We can see the architecture underneath. The tradeoffs. The failure modes. The gap between a solution that works today and one that holds at scale.

Key Takeaways

  • Dialogue requires presence: issuing commands isn't engineering, conducting Socratic interrogation is
  • Ownership remains personal: distributed authorship doesn't mean diffused responsibility
  • Taste is learnable but real: your instinct for good design improves over time, and it's yours to keep
  • Judgment is the differentiator: when anyone can generate anything, knowing what to build matters more than building it

The Bottom Line

The code got easier. The questions got much harder. This isn't about accepting mediocrity — it's about understanding that the skill ceiling has moved. Engineers who treat AI as a dialogue partner, take ownership of outcomes they can't fully explain, and cultivate genuine taste will be fine. Those treating it like an auto-mechanic will find themselves explaining to HR why they're replaceable.