The development world as we know it is at a turning point. It's been at a turning point a lot of times since I've started doing this some 16 years ago, and a lot more times before that. But this one time is special, this one time is different: AI threatens to replace us. Get out of Starbucks and learn to use a shovel.
Relax, I'm dramatic on purpose, we're not going anywhere. I'm writing this post mostly for younger folks who I don't envy, because they're in a position where they are threatened by, scared by, and forced into the AI bubble. People are losing their minds over whether AI will make them obsolete, and they're listening to false prophets who are telling them that they should be learning a real craft, a physical skill that won't be touched by AI any time soon. But what's the reality? Reality is that AI is a tool (last time I checked). A powerful tool, and it could be a great asset to a developer who knows what they're doing. If you don't know about the basics, security, user experience, performance... no amount of AI will help you make a good app. That's the truth, and anyone who's telling you otherwise is likely fearmongering to raise their importance, or trying to sell you an AI-powered service or one of their dime-a-dozen courses. These people are not your friends!
I'm not saying you need to avoid AI until it disappears. On the contrary, it's here to stay, just maybe not in the areas AI doom merchants want you to believe it will. You should definitely learn how to work with AI. Large Language Models (LLMs) changed the game for me: I can get to the prototype faster, I can debug faster. I abstracted away boring multi-file edits, tests, and boilerplate, which are generally time consuming. Do I work more? No. Do I output more? Yes, but not as much as my CEO would like me to. The time I save with AI, I spend prompting that very same AI about the stuff I know the least. Brainstorming, asking it to explain its choices, then cross-referencing against documentation and people who actually know what they're talking about. More often than the doom merchants would like to admit, its choices turn out to be overkill or outright wrong. That's how I'm filling the gaps in my infrastructure and backend knowledge.
It's easy to pay $200+ a month and have Claude or any agent write the app for you, but what happens when you get a data breach and you don't understand the code well enough to fix it? When you can't explain to your users what went wrong because the AI made decisions you never questioned? Will you be able to sell your service when everyone can use that same $200+ subscription to build their own, with the same bugs, the same blind spots, and the same confident nonsense baked in? Is that sustainable?
I don't think so.
"But Mythos is coming! ChatGPT 6! $NEXT_BIG_THING! It's going to obsolete us!"
Good. Let it. Add it to the toolset so it can help you learn, not just crunch code faster.
So, what do you actually do?
If you're a junior engineer and the section above made your stomach turn, good, that was the point. Now take a breath.
Keep your ears to the ground on AI. Keep your hands on the fundamentals.
Use whatever agent you can get your hands on and get a feel for it. That's it. That's the whole strategy. Everything below is just me explaining why it works.
Pay attention, don't panic
You should know what the current models can and can't do. Try the tools. Read the release notes. Notice which parts of your job got easier this quarter and which parts got weirder. That's "ears to the ground", it's awareness, not obsession. You don't need to try every new agent framework that trended on Hacker News last Tuesday. Trust me, I lived through JavaScript framework fatigue: the industry always converges on a handful of choices that actually work, and the rest quietly disappear into blog post graveyards. Most of them will be gone by Tuesday after next, and the ones that survive will still be around when you get to them.
The people sprinting to learn the agent-of-the-day are running on a treadmill that someone else keeps speeding up. Don't join them. Walk. Pick one agent, get comfortable with it, and stop worrying about the rest. The core concepts carry over: context windows, tool calls, prompts, rules files, how the thing behaves when it's unsure. Switching from one agent to another is a weekend of friction at worst, not a career setback.
Learn the thing underneath
Whatever the AI writes for you sits on top of something older and more stable: HTTP, SQL, a filesystem, a process model, a type system, a cache, a queue. When the AI produces something broken (and it will), your ability to fix it depends entirely on whether you understand the layer it was writing into.
So the unglamorous advice: learn how a request actually travels from a browser to your database and back. Learn what an index is and why your query is slow without one. Learn what happens when two users hit the same endpoint at the same time. Learn why your deploy broke at 2 AM. None of this is going out of style. None of it is getting abstracted away, no matter what the keynote says. And yes, it sounds like a lot, but every hour you put into this now pays off twice over once you pair it with AI.
If you know these things, AI becomes a lever. If you don't, it becomes a very confident liability with your name on the commit.
Choose boring tech that works
Here's the part that won't trend anywhere. The best thing you can build your career on is boring technology, the stuff that's been around long enough to be unexciting. Postgres. A cron job. A plain old background worker. SQLite on a single box, serving more traffic than you'd believe. A monolith that fits in one repo and one head.
Boring tech is boring because it works. It has documentation written by humans who used it in anger. It has failure modes that are known, named, and googleable. It doesn't rename itself every six months. And here's the quiet part: AI is dramatically better at boring tech than it is at the shiny stuff, because there are decades of examples to learn from. Using AI on top of a boring stack is where it actually shines.
Meanwhile, the person who built their whole identity on a framework that launched last spring will spend the next year explaining why nothing works, and why nobody can hire for it.
The job is still the job
Users still need software that solves their problem without asking them to set up ten cloud services, babysit backups, or pray that availability holds. Businesses still need people who can look at a system, understand it end to end, and make a judgment call. AI doesn't do judgment. It does confident averages. Sometimes dangerous ones. Judgment is yours to build, and you build it by doing the work: reading code, breaking things, fixing them, asking why. And breaking things again, just to be sure.
Use AI to skip the parts that were always busywork. Use the time it gives you back to get better at the parts that aren't.
Will our lives change? You bet they will. Which way, that's up to you. You need to decide if you'll chase the agent-of-the-day or you'll actually use AI to produce boring solutions that actually work.
Your call.
I got this figured out for myself. I think I'll still have a lifetime of work fixing vibe-coded messes false prophets will inevitably create. It'll likely be boring but, sometimes, boring is good.
