Career, Hiring, and AI Advice
- https://www.youtube.com/watch?v=cSIMVYjVF28
- Original title: The Pragmatic Engineer AMA
Gergely Orosz answers subscriber-submitted questions (asked by guest host Volodymyr Giginiak, CTO of Wordsmith AI) in a reversed-format AMA episode of the Pragmatic Engineer podcast, covering why he left Uber to write full-time, how AI is reshaping hiring and interviewing, which engineers are most in demand, how big tech companies differ in AI adoption, the quality-versus-speed tradeoff in AI-assisted coding, engineering-manager expectations, career advice across markets and seniority levels, and the economics of running an independent newsletter business.
Leaving Uber to write full-time
Gergely was an IC-turned-manager at Uber for about four years. After 2020 COVID layoffs hit his team and morale, he realized he no longer enjoyed being a "middle manager" — keeping both his team and his own management chain happy through politics and reorganizations. His original plan was to leave Uber, finish The Software Engineer's Guidebook in six months, then start or join a startup, funded by ~$400k in vested Uber stock. Writing the book took far longer than planned (a classic "first migration always takes longer than you think" trap). While weighing whether to raise money for a startup (his idea: productize Uber's internal RFC review tooling), he realized his real motivations were money and the feeling of being on a small, focused team — not genuine excitement about the startup idea itself. Inspired by Lenny Rachitsky's early Substack numbers, he decided to try a paid newsletter for software engineers instead, reasoning there were far more engineers than product managers and no one was serving them with a paid publication yet. It took off within weeks.
Hiring and interviewing in the AI era
Pre-AI hiring had two dominant models: Google-style algorithmic/LeetCode interviews (selected for CS fundamentals, grace under pressure, and tolerance for corporate process) and startup-style trial weeks or take-home tasks (selected for practical ability). AI has broken take-homes as a signal, since candidates can now complete them near-perfectly with AI help. Gergely expects both models to persist but shift: filtering will lean more on AI-assisted take-homes followed by in-person/whiteboard rounds where AI isn't available, and companies like Wordsmith are redesigning interviews to probe whether a candidate can reason about and correct what AI produces, not just accept its output. Linear's model — working alongside the team for a real week — is a good practical filter but doesn't scale, since only well-known companies can get candidates to take the time off. Overall, Gergely expects hiring to become messier, more subjective, and to feel more unfair as old clear rules stop applying.
Who is in demand right now
Engineers thriving today are "product-minded," don't stay in their narrow lane, and got early hands-on experience building things with AI (RAG, fine-tuning, choosing off-the-shelf vs. custom models, inference cost tradeoffs) — essentially the "cloud engineer" of five years ago, now applied to AI infrastructure. People stuck at companies with no AI exposure or without a strong company pedigree are struggling to move up a tier. Gergely's informal tier model: consulting shops (struggling) → product companies → venture-funded product companies → AI labs (Anthropic, OpenAI) at the very top, which are now extremely selective. Lower-level/systems/embedded engineering seems comparatively less saturated and less AI-automatable so far. For juniors, pedigree (school, internships) or visible side projects/open-source contributions matter more than ever, and taking any job over no job — then excelling there — is sound advice.
How different companies are handling AI adoption
Anthropic (via Claude Code) is the closest example of "AI-native SDLC": prototypes replace design docs, and the product (Claude Code) is really a byproduct of being an AI research lab first. Gergely doubts this is broadly replicable outside of an AI lab. Among big tech: Google is investing broadly and has the advantage of owning a real model (Gemini) but is chaotic internally; Meta is bogged down chasing its own model and morale is falling as engineers are reassigned to AI data labeling; Microsoft is hampered by internal politics between Copilot and Core AI orgs and struggles with reliability; Apple is secretive with what he describes as weak internal engineering culture, though its focus on device/software basics may still pay off; Amazon has built its own (subpar) internal AI tooling (Kiro) but engineers default to Claude Code anyway, illustrating how hard it is to retrofit AI into a large, established org. Smaller public "little tech" companies (Uber, Ramp, Intercom) are doing comparatively better because they don't try to own the whole AI stack — they just adopt the best available tools and optimize their business. Uber, for instance, has built extensive in-house AI tooling (agent builder/studio, an internal CLI, background "Minion" agents, AI-aware code review and on-call routing) despite having only ~3,000 engineers.
Quality versus speed, and standards
Gergely rejects the idea that code quality must be sacrificed for AI-driven speed as a false dichotomy — it's really about which stage of a product you're in (borrowing Kent Beck's explore/expand/extend framing): early-stage prototypes can tolerate hacks, scaling products need more care, and mature/revenue-critical products need real discipline. He cites Uber's old, wasteful "poll every 5 seconds" mobile architecture as pre-AI evidence that tech debt has always been a speed/quality tradeoff, not something new. He notes AI also makes refactoring and technical-debt cleanup much cheaper, so there's less excuse not to do it. On emergent standards (e.g., MCP), he thinks they arise accidentally from smaller, "non-threatening" labs rather than being planned — a dominant lab proposing a standard today would be seen as a lock-in play.
Engineering managers, productivity, and the pendulum
The industry (following Twitter/X's post-acquisition model) is currently pushing engineering managers to code more and focus less on people management, meaning less support and championing for individual engineers in the near term; Gergely expects this to be a pendulum that swings back. On measuring "AI productivity gains" separate from just more code and PRs, he argues it only really shows up as either incremental revenue from new AI-powered products or as cost savings — and he suspects AI's biggest realized value so far is cost savings, comparing its trajectory to cloud computing (ubiquitous infrastructure, not a new market-creating technology like mobile was). He also warns against believing AI simply makes engineering easier — used well, it should make you think as hard or harder, not less.
Career advice across situations
For engineers worried about relevance (a QA engineer in banking, a self-taught engineer struggling post-layoff, a student going into game development), his consistent advice is: get hands-on AI experience inside your current job/company if at all possible (propose a small experimental AI project) rather than quitting for a full CS degree or waiting for side-project motivation to strike; degrees matter more now than during the 2015–2020 bootcamp-hiring window, partly for filtering and partly for visas; and building a real network of motivated peers (changing teams, companies, or communities) matters more than raw individual effort. For EU engineers in the current market, he recommends staying opportunistic — job openings have picked up compared to the brutal 2023 market, and taking interviews doesn't commit you to leaving.
Personal workflow and the newsletter business
Gergely writes without AI assistance — he finds AI-generated drafts don't sound like him and that the thinking, not the typing, is the valuable part of writing. He does use AI heavily for research (deep-research tools) and is building his own backend (subscriber management, support tooling) using Codex, Claude Code, and Cursor. He shared that The Pragmatic Engineer newsletter now has 10,000+ paying subscribers (grew from 100 paying subscribers in week one to matching his best Uber total compensation within four to five months), and that he deliberately doesn't paywall everything even though it would grow revenue further. He recounted two brushes with legal trouble over investigative articles (an unpublished piece on Bunq, and a published, heavily fact-checked deep dive on an events-company startup that led to related BBC reporting) and said investigative journalism ultimately isn't for him. He closed by predicting that in five years there will still be strong demand for engineers who are genuine craft professionals — people without ego who choose the right tool for the job and understand tradeoffs — regardless of how much the tooling itself changes.