What to Actually Learn at Each Stage of an Engineering Career (in the AI Era)

A new subscriber found this newsletter last week the best possible way: he replied to a post. He's a brand-new data engineer, and his question was the one almost every engineer is quietly asking right now, no matter their stack: what should I actually learn?

It's a harder question than it used to be. Not because there's too little to learn, but because there's too much, and a model on your other screen will happily generate ten more things to learn before lunch. The skill that matters most now isn't any single tool. It's knowing what to learn at your stage, and having the nerve to ignore the rest.

So here's the map I wish someone had handed me. It's organized by rung, not by stack. A data engineer, a backend engineer, and an SRE hit the same walls in the same order. The doorways look different. The staircase is identical.

Early career: build a spine, not a toolbox

The trap at the start is collecting tools. A new framework every week, a new database every month, a tutorial backlog you'll never clear. It feels like progress. It's mostly motion.

What you actually need at this stage is a spine: a small set of fundamentals deep enough that everything else hangs off them.

• One language, actually deep. Not five languages at tutorial depth. One you can reason about when it's 2am and the stack trace makes no sense.
• Data and how it's modeled. SQL, schema design, and what happens to both at scale. This is universal. Whether you call yourself a data engineer or not, your career runs on data that's shaped well or shaped badly.
• How systems fail. Not just how they work. The engineer who understands timeouts, retries, and what a queue does when it backs up is worth three who only know the happy path.
• Reading more code than you write. Your fastest growth is in the codebase you already work in, not in a new side project.

The AI-era twist: a model can hand you working code before you understand it, which means you can now ship things you can't debug. That's the new failure mode for early-career engineers. So use AI to go faster on things you already understand, and to explain things you don't, but never to skip the understanding. The goal of this stage isn't output. It's becoming someone who can tell when the output is wrong.

Mid-level: learn to own the ambiguous thing

You become mid-level the day you stop needing the ticket fully specified. The wall here is ownership. Junior work is "here's the task, go do it." Mid-level work is "here's the rough problem, figure out the task."

What to learn:

• Decomposition. Taking a vague problem and breaking it into pieces small enough to actually start. This is the single most underrated engineering skill, and almost nobody teaches it directly.
• Tradeoffs over correctness. There's rarely one right answer, only answers with different costs. Start naming them out loud: faster but harder to maintain, cheaper but slower, simple now but a ceiling later.
• Communication as a deliverable. A clear update, a tight PR description, a design doc someone can actually follow. Your code stops being judged in isolation and starts being judged by how well others can build on it.

The AI-era twist: AI is fastest at exactly the work that used to fill a mid-level engineer's week, the well-specified, medium-hard chunk. So the value moves up a layer, to framing the problem and judging the result. Lean into the judgment. The engineers who treat AI as a faster typist will plateau. The ones who treat it as a fast, tireless, sometimes-wrong collaborator they have to direct will pull ahead.

Senior: trade output for leverage

This is the rung where the rules quietly change, and most people don't notice. Senior is the last level you reach almost entirely on personal output. Every level above it is measured by what you make possible, not what you produce.

What to learn:

• Leverage over heroics. The design decision that makes ten tickets easier beats the heroic ticket you closed at midnight. Same hour, an order of magnitude more reach.
• Judgment that changes what the team does. Being reliably right about what matters, what to build, what to cut, which risk is real, so your read on a problem actually moves the plan.
• Growing other people. Give away the high-visibility work that made you look good and coach someone through it. You trade a little spotlight for the one thing the next level requires: evidence you make others better.

The AI-era twist: when everyone on the team has a model that can produce a competent first draft of almost anything, raw production gets cheap and taste gets expensive. Knowing what's worth building, what's good enough, and what's quietly wrong is the senior skill the AI era pays the most for.

This is the rung the Top Engineer Method is built for, the climb from strong senior to Staff. If that's the wall you're standing at, that's your on-ramp. If you're earlier, file it away for later. Honesty now beats a sale today.

Staff and principal: think in systems and bets

The wall at Staff isn't technical depth. You have that. It's scope. You stop optimizing a service and start optimizing how a whole org builds.

What to learn:

• Systems thinking past the codebase. Org structure, incentives, where the real bottleneck is (it's usually not the code). The Staff engineer who can see that two teams are duplicating work is worth more than one who can shave a millisecond.
• Influence without authority. You'll drive things across teams that don't report to you. That's persuasion, written strategy, and trust, not a title.
• Placing bets. Deciding which few things matter this year and defending that focus when everyone wants everything. Increasingly, that includes the honest call on where AI changes the architecture and where it's a distraction.

The AI-era twist: AI raises the ceiling on what a small team can ship, which makes direction the scarce resource. A team pointed at the wrong thing now gets there faster. Staff engineers who set direction well are about to be worth a great deal more.

Leadership: your output is the team

Some engineers step off the technical ladder onto the management one. It is not a promotion so much as a career change that happens to share a hallway. The wall is total: your old scorecard, what you shipped, gets deleted. Your new one is what your team becomes.

What to learn:

• People as the system. Hiring, growing, and yes, sometimes losing people. The leverage that used to come from a good abstraction now comes from a good team.
• Clarity at scale. Setting context so thirty people make good decisions without you in the room. Most management failures are clarity failures.
• Protecting focus. Absorbing organizational chaos so your team can do deep work. Increasingly that means having a real answer to "how should we be using AI here," instead of either banning it or mandating it.

The AI-era twist: as AI compresses execution time, the gap between teams is less about raw speed and more about whether they're aimed well and growing. That's leadership. The job gets more human as the tools get more capable, not less.

This rung is what the Engineering Leader Method is for, the move from senior IC or new manager into someone the org is built around. Same rule as before: it's for you when you're reaching for that wall, not before.

The one skill underneath all of them

Read back through the rungs and there's a thread: at every stage, the real skill is choosing what to learn and what to ignore. It's just that the unit gets bigger. Early on you're choosing which language to go deep on. By the top you're choosing which bets an org makes.

The AI era turned up the volume on this until it's the whole game. The constraint was never access to information; it's been a flood for years. The constraint is judgment about what deserves your attention. A model will generate infinite things to learn. Knowing which three matter, for the rung you're on right now, is the meta-skill the whole career runs on.

So to the data engineer who kicked this off, and to everyone else reading: don't ask "what's the hot tool." Ask "what does my next rung actually reward," then learn that, deeply, and let the rest go. The staircase is the same for all of us. You just have to climb it on purpose.