What compounds when the agents do the typing.
"The tools we use have a profound (and devious!) influence on our thinking habits, and therefore on our thinking abilities." — Edsger Dijkstra
Every wave of automation follows the same pattern: the craft moves up a level. Compilers didn't kill programming; they made assembly-language craftsmanship rare and high-level abstraction universal. Version control, package managers, cloud services, container orchestration, managed databases — each absorbed a layer of skilled labor and freed engineers to work on a harder problem.
Agents are the next wave, and they absorb a bigger layer at once. That is scary and it is also ordinary. What will not change is the thing that makes a good engineer good in the first place: taste, judgment, responsibility, and systems thinking.
quadrantChart
title Skills in an AI-augmented career
x-axis Decays with AI --> Compounds with AI
y-axis Low leverage --> High leverage
quadrant-1 Invest heavily
quadrant-2 Delegate to AI
quadrant-3 Safely ignore
quadrant-4 Keep as muscle
Rote syntax recall: [0.05, 0.15]
Boilerplate writing: [0.10, 0.20]
Keeping up with framework churn: [0.20, 0.40]
Reading diffs carefully: [0.85, 0.85]
System design: [0.95, 0.95]
Debugging distributed systems: [0.80, 0.85]
Writing clear specs: [0.90, 0.90]
Product taste: [0.90, 0.85]
Mentoring: [0.75, 0.75]
Memorizing LeetCode: [0.10, 0.30]
Skills whose value goes up when AI gets better:
Skills whose value goes down (but don't hit zero):
These are not useless — you still need some of them — but they shouldn't be the center of your identity as an engineer anymore.
A few principles that will, I think, hold up.
An agent is a junior engineer that can type at superhuman speed but can't be woken up at 3 a.m. to own a Sev-1. Your job is to give it instructions, set the boundaries, and take responsibility for the outcome. That is senior work. Do it.
Read every diff. Read every paper you think you should. Read the docs you're tempted to skim. Read the customer feedback. Read competitor postmortems. The compounding effect of reading is unchanged — maybe amplified — by AI.
A blog. A newsletter. A public repo. Internal docs your team actually uses. Writing is how you turn private understanding into public leverage. Agents will summarize the world; your writing is the corpus the next generation trains on. Your voice matters.
Every chapter in this handbook pointed at the same thing: ship, measure, iterate. The AI era rewards this more, not less. The feedback loop from "idea" to "live system" is shorter than it has ever been. Use that.
The field is full of confident predictions and most of them age poorly. Say "I don't know" when you don't. Update when the evidence changes. Hold your strongest opinions loosely. This is how you stay trustworthy as the landscape keeps moving.
Your brain is the scarce resource. Sleep. Walk. Cook. Read fiction. The best engineering ideas still come in the shower, on a hike, or at 2 a.m. on a quiet Tuesday — and no amount of AI will generate them for you.
I don't know, and nobody does, exactly where this goes. The plausible range runs from "a very powerful tool that changes a lot of jobs" to "a civilization-scale transition." All of those futures are being built, right now, by engineers and researchers who are no smarter than you, using tools that are getting better every month.
You get to have a hand in what kind of future this is. The decisions about what to build, what to refuse to build, what standards to set, what culture to bring to your team — these are not decisions policymakers or CEOs make alone. They are decisions engineers make in a thousand small pull requests every day. Make yours thoughtfully.
You picked a good time to be a backend engineer. The tools are better than they have ever been. The bar is rising, and the rewards for crossing it are bigger than they have ever been. The story is still being written.
Go build.