A.I. will make the bad "badder", the rich richer, the poor poorer, the hopeful more hopeful, the doubtful more doubtful. It's given people the ability to do more with less—and less with more. It is the double-edged sword that is both blessing and curse.
The question isn't whether A.I. will be good or bad for humanity. It'll be both. We know this.
The real question is: How much of a force multiplier will it be on the things that matter most? And where will that take us?
What I'm Seeing Out There 📊
Let me be honest about what's happening in our industry right now.
Junior developer roles have collapsed. Entry-level tech postings fell 60% between 2022-2024, and Big Tech hired only 7% new grads in 2024. At the same time, the barrier to entry for creating has never been lower. Suno has 12+ million users making music. Over 15 billion AI images have been generated. GitHub Copilot users complete tasks 55% faster.
A.I. makes mistakes AND solves age-old math problems. It takes jobs and creates some. It's used to improve breast cancer detection throughput by 44%, and it's used in warfare and surveillance.
Among all this -- it is the end of software developers as we knew them—and the start of something new: The Software Creator.
❓ "Should I Still Learn Software Development?"
Short answer: Yes. These tools are not bulletproof. Not even close.
What I've Learned About A.I. Capabilities
I think of A.I. agents like employees I'm onboarding. Right now, A.I. performs at junior-to-mid level for routine, well-defined tasks—but falls well below junior on complex system changes and architectural decisions.
| Task Type | AI Performance |
|---|---|
| Single functions (HumanEval) | 90-96% ✅ |
| Standard GitHub issues | 70-76% 🟡 |
| Complex real-world problems | 17-23% ❌ |
The breadth of knowledge is WIDE, but not DEEP. I learned this the hard way building an AI-powered chatbot for my team—the thing could answer simple questions beautifully, but the moment someone asked something that required understanding our specific infrastructure (or a tool call gave bad info), it would confidently hallucinate an answer that sounded perfect but was completely wrong.
Additionally if you don't know how to ask, "WHY are you doing it that way?" or better yet, "WHOA, please don't do that, do this instead," you'll get taken down a rabbit hole of confident failure.
The Real Productivity Numbers (They're Messier Than You Think)
You've probably heard "AI gets you 80% there." The data tells a more honest story.
GitHub Copilot generates 46% of code for active users, but the acceptance rate averages only 30%—meaning 70% of suggestions get rejected. 66% of developers say their top frustration is "AI solutions that are almost right, but not quite." And here's the kicker: a METR randomized controlled trial found developers actually took 19% longer with AI—while believing they were 20% faster. 🤯
I've felt this myself. There's something intoxicating about watching code appear on your screen. You feel productive. But then you spend an hour debugging something the AI introduced three suggestions ago, and you realize your "80% there" was actually "60% there with new bugs."
Realistic ratio: 60/40 or 70/30. That "last mile" is longer than it looks.
🛠️ What Should You Actually Learn?
You don't need to know everything overnight. I didn't. I came into this field without a computer science degree and learned everything on the job over 11+ years. Buy tools for your garage as you need them—same principle here.
Start With the Foundations
Learn the basics first: variables, functions, classes. These are universal to any modern language. Here's my unpopular opinion: build something simple without A.I. first. That trial-and-error is what makes you stronger. It's what lets you recognize when the AI is leading you astray.
And learn Git. Non-negotiable. This will be important for keeping your code and the A.I.'s code on track.
Then Build Your Stack
Backend fundamentals matter—Firebase, Express/Node, Flask. Understand what an API is and how to stand one up. If you're working with AI agents, TypeScript + Zod (or Pydantic for Python folks) become critical. Type definitions and data validation are how you keep AI-generated code from doing something unexpected at runtime.
Pick up basic testing with Jest or pytest. Here's a common trick I practice: make the AI write tests, then make it run them. It's a forcing function for quality.
Security basics should be non-negotiable too. Don't put secrets in frontend code (I've seen this more than I'd like to admit). Understand OAuth and auth flows. Use a secret manager.
Level Up When You're Ready
Cross-platform frameworks like Electron, Capacitor, and Quasar open up new possibilities when you're ready for them. Docker is technically a "bonus" but honestly, just learn it. Your future self will thank you.
💼 "How Do I Get a Job?"
If You're Starting from Zero
The barrier for entry is higher now. I won't sugarcoat that. But I've hired people, and I can tell you what actually moves the needle.
First, build a real portfolio WITH A.I. Release software on GitHub that's actually useful. Demonstrate you can go from idea to reality. But here's the crucial part: understand it under the hood and be able to explain it. I've interviewed candidates who built impressive-looking projects but couldn't explain basic architectural decisions. That's a red flag.
Second, bring genuine enthusiasm. I'll take someone who's green but humble and eager over someone excellent but difficult to work with every single time. Your soft skills matter more than ever in a world where AI can write code. The human stuff—collaboration, communication, judgment—that's your edge.
Third, develop AI skills beyond basic prompting. Job postings for "prompt engineers" peaked in April 2023 and have plateaued since. Basic prompting is table stakes now. What employers actually want is system design and architecture, MLOps and production deployment, LLM fine-tuning and RAG development, and critical evaluation of AI outputs.
Jobs mentioning AI skills pay a 28% premium (~$18K/year), rising to 43% for multiple AI skills.
If You're Mid-Level or Beyond
Move into management because those jobs are safer. Right???
Nah. The data actually shows middle managers comprised 32% of 2023 layoffs, and companies like Microsoft, Amazon, and Meta are explicitly flattening org structures. "The Great Flattening" is real. I say this as someone in management—the title alone doesn't protect you.
What actually works is becoming a force multiplier. Use A.I. to do an amazing job at what your employer cares about. I've built reports and automations that would have taken weeks and compressed them into days. That's the kind of thing that gets noticed.
Learn the tools deeply. Explore, play around, but focus on what connects to your actual job. Build with A.I. agents—learn how to orchestrate them, multitask with them, and free up your time for the work that actually requires human judgment.
Lead without the title. Technical design, mentorship, and judgment are human superpowers. No AI is going to sit down with a frustrated junior dev and help them work through impostor syndrome.
This is what transforms you from Software Developer → Software Creator → A.I. Orchestrator.
🎮 How I Got Started with A.I.
Here's roughly how my journey went:
| Phase | What I Learned |
|---|---|
| 🎨 Image Gen | Started with Stable Diffusion. Check out Flux, Imagen, Midjourney |
| 💬 Chatbots | ChatGPT, Claude, Gemini. Learned context windows, tokens, prompting |
| 🤖 Coding Agents | Cursor, Windsurf, Claude Code—AI edits code locally |
| 🔧 Building Agents | CrewAI, LangChain, MCP servers |
| 🔬 Deep Research | Gemini Deep Research, NotebookLM for studying |
| 🏠 Running Local | Ollama, vLLM, fine-tuning your own models, RAG |
Take it one step at a time. I spent months in each phase before moving to the next. You don't need all of this at once.
⚠️ Real Talk: What's Coming
I'd be crazy to tell you not to be concerned. A lot is changing. But here's how I think about it:
A.I. amplifies what people already do—good and bad. People generally want everyone to be well off (so they can be well off too). And people don't like being squashed or oppressed.
We're going to figure things out, as we have before. But it'll take work.
What About "The Next Big Thing"?
You might hear about transformer alternatives, neurosymbolic AI, quantum computing integration. Here's my honest read on these:
Transformer alternatives like hybrid models exist (Jamba, Granite 4.0), but transformers remain central. We're seeing augmentation, not replacement. Neurosymbolic AI produced impressive results with AlphaGeometry, but no major LLM has shipped a general product yet. And quantum + AI? Jensen Huang estimates 15-30 years for "very useful" applications.
These are research directions to watch, not imminent paradigm shifts. Focus on what works today.
🎯 The Bottom Line
Yes, still learn software development. It's still worthwhile.
But understand this: the role is transforming. The developers who thrive will be the ones who become creators—people who can direct A.I. tools with expertise, know when to trust output and when to override, build entire systems (not just write code), and amplify their output through orchestration.
Be kind to each other, and be kind to yourself. Be curious.
The future will be great and problematic. You need to know how to spot deepfakes, understand security risks, and leverage the tools—otherwise you'll be pushed out.
But you've got this. 🤝
What's your experience with A.I. in your development workflow? Are you a skeptic, an enthusiast, or somewhere in between? Drop a comment below!
Top comments (0)