The New Bar for Engineers in the Age of AI
TL;DR
The new bar for engineers is product judgment, not just coding skill — Gergely Orosz says the standout engineers at Uber weren’t just strong coders; they were “product-minded,” talking to PMs, understanding business metrics, and cutting months of engineering work by questioning whether a feature detail really mattered.
AI compressed years of coding advantage into months — Orosz describes genuine “grief” that skills which took 10+ years to build are now widely accessible, saying the shift from autocomplete to AI writing code as well as he could happened between roughly September and December.
Managers are being squeezed as teams get smaller and flatter — since 2023, he’s seen the classic 5–10 person “two-pizza team” shrink toward 2–4 people, with fewer engineering manager roles and more demand for technical leaders who can still “dive deep.”
Uber’s famous microservices story was mostly misunderstood by the industry — the company’s thousands of services weren’t cargo-cult architecture; they were a scaling strategy to let mostly junior teams ship independently while senior engineers concentrated in platform teams.
Expertise still matters because AI is persuasive, not accountable — he illustrates this with a legal example where ChatGPT, Claude, and Perplexity all gave the same confident but wrong answer, while a real lawyer explained the regulation is interpreted differently in practice.
Career myths are breaking too: empty GitHubs can hide great engineers — Orosz argues some of the best hires are the “boring” ones who leave at 6 p.m., pick up their kids, and do eight focused hours of excellent work, rather than broadcasting passion through side projects and open source.
The Breakdown
From PHP tinkering to seeing why Nvidia won
Gergely Orosz opens with a very unromantic origin story: he got into coding through “hacking” with PHP and JavaScript in high school, then nearly lost interest in computer science because university buried him in differential equations, probability, physics, C memory allocation, and 3D graphics. In hindsight, he says those fundamentals paid off — when LLMs took off, he could intuit why Nvidia mattered because GPUs, matrix transformations, and parallelization all clicked from those old graphics classes.
Uber at peak hype: duct tape, Google Sheets, and 5,000 microservices
His Uber years were exactly the contradiction people imagine: from the outside it looked like the hottest company on earth in 2016; inside, “everything is held together by duct tape and Google Sheets.” The memorable insight here is his explanation of Uber’s microservices explosion: the company knew it would go from roughly 200 to 2,000 engineers and mostly hire junior talent, so senior engineers built platform infrastructure while independent teams shipped their own services — a strategy that worked brilliantly for Uber but got copied out of context by everyone else.
The accidental manager who called the project red
Orosz’s move into management happened because he was the only person blunt enough to say a rewrite project in Amsterdam was not green, despite status meetings marking everything green week after week. He told his manager the team needed demos, pairing, stronger ownership on critical work, and feature cuts; a month and a half later, leadership changed and he was asked to run the project precisely because he had named the problem and proposed fixes. His takeaway is practical: you often don’t know if you’ll be good at management until you try it in a reversible setup, like Uber’s six-month apprentice manager program.
What separates engineers who compound from those who plateau
The strongest engineers he saw were not just technically solid — they were deeply curious about what the team and the business were actually trying to do. They set up one-on-ones with PMs, watched competitors, jumped into code reviews, and in planning meetings could challenge requirements in a way that saved huge effort, like pointing out one backend-dependent state change would turn a few weeks into a few months of work. His label for them, years before it became fashionable, was “product-minded software engineers.”
The grief of watching coding become abundant
This is the emotional center of the interview: Orosz says it “sucks” that coding, which used to be a hard-won signal of grit and intelligence, is becoming something “everyone can do” with AI. He describes spending years getting better at debugging, switching stacks, passing coding interviews, and shipping to production — only to feel that the value of that hard-earned edge collapsed in a few months once newer models started producing code at or above his level, especially in unfamiliar frameworks.
Why AI raises the premium on real expertise
He pushes back on the idea that AI erases learning, using a legal anecdote that lands hard: ChatGPT, Claude, and Perplexity all confidently told him outcome A, while a lawyer later said the real answer was B because courts don’t interpret the regulation the way the models assumed. That, to him, is the pattern — novices overtrust, experts catch the gaps. Software will likely work the same way: even if Claude can generate code, businesses still want accountable engineers who can explain failures, own outcomes, and prevent the next incident.
Smaller teams, fewer managers, and career myths getting exposed
On careers, Orosz says two shifts are already visible: teams are shrinking from the old 5–10 person “two-pizza team” to more like 2–4 people, and engineering management is becoming less common and less lucrative, with some companies consolidating 30 engineers under one director plus tech leads. He also punctures two myths: that frequent job-hopping is always the best strategy, and that the best engineers are the loud ones with packed GitHubs — often the best hires are the “boring” engineers with empty profiles who quietly do eight hours of focused work and outperform the flashy crowd.
The old-timer trap, Boris Cherny, and the next five years
His closing frame is Grady Booch’s analogy that AI is another “leap of abstraction,” like the move from assembly to higher-level languages. The winners weren’t the people clinging to the old way; they were the ones willing to learn the new abstraction without ego. That’s why he points to Anthropic’s Boris Cherny — a strong engineer who once switched toward functional programming after a motorcycle crash left him typing with two fingers, and now uses Claude to generate nearly 100% of his code — as the template: adaptable, curious, and happy to let go of old identities.