I Learned to Code in the AI Era — Here's What I've Learned About Experience

· 7 min read
#ai-coding#paradigm-shift#career#productivity

In late 2022, ChatGPT launched. I was a product manager with some technical background but no real coding skills. I’d dabbled, sure. Written some scripts. But I wasn’t a developer.

By 2025, I’ve shipped traditional software, built AI agents, created an entire coding course, and run systems with 10+ parallel AI agents building features simultaneously.

I did this without learning to code the traditional way. And that gave me an advantage — at first.

But after three years of working this way, I’ve realized something: experienced engineers have the biggest upside. They just have to learn how to unlock it.

Why Fresh Eyes Helped Me Start Fast

When I started coding with AI, I had no preconceptions about what the workflow “should” look like. I didn’t know that “real developers” write their own code. I didn’t know that spending an hour on specs before writing any code was “inefficient.”

I just did what worked.

And what worked was:

  • Spend more time on context than code
  • Spec out what I wanted before asking AI to build it
  • Treat the AI like a smart but uninformed collaborator
  • Validate output against specs, not by reading every line

I didn’t have muscle memory fighting against me. No instinct to just start typing. For me, writing specs and context was never overhead — it was the only way I knew how to work.

That’s why beginners can sometimes pick this up faster than veterans. No habits to unlearn.

But here’s what I’ve come to understand: starting fast isn’t the same as going far.

The Secret Nobody Talks About

You can only create templates as good as your engineering.

This is the part most people miss — including a lot of AI hype bros predicting the end of developers.

AI is a force multiplier. If you multiply zero by ten, you get zero. If you multiply 15 years of hard-won patterns, architectural knowledge, and battle-tested intuition by ten? That’s where it gets interesting.

When I template a workflow, I’m encoding what I know — which is limited. I can spec out a feature. I can describe what I want.

When a senior engineer templates their workflows, they’re encoding:

  • Patterns that work at scale (and knowing which don’t)
  • Edge cases they’ve been burned by before
  • Architecture decisions that prevent pain later
  • Code review instincts built over thousands of PRs
  • The taste that comes from shipping real products

A senior engineer can spec out a system — one that doesn’t fall apart when you add the second feature, or the tenth, or the hundredth. That’s a different ceiling entirely.

Engineering Isn’t Dead — It’s More Valuable Than Ever

The “developers are obsolete” takes are wrong. Completely wrong.

What’s changing is the leverage, not the value of the skill.

We still need people who:

  • Know what to build and why
  • Can evaluate whether AI output actually solves the problem
  • Understand systems deeply enough to template them well
  • Can debug when things go wrong (and they will)
  • Have the taste to know good from bad

A great engineer with AI tools can do the work of a team. A mediocre engineer with AI tools can do the work of… a slightly faster mediocre engineer.

The gap between good and great is widening, not shrinking.

Template Your Engineering

This is what I’ve learned from working with experienced developers who actually get it:

The ones who thrive don’t fight AI. They template themselves.

They take the patterns in their head — the ones built over years of shipping real software — and they encode them:

In CLAUDE.md files:

“We always use repository pattern for data access. See src/repositories/ for examples. Never put database calls directly in handlers.”

In spec templates:

“Every feature spec must include: user story, acceptance criteria, edge cases, error states, and rollback plan.”

In custom commands:

“When I say /review, run this code review checklist that catches the 20 issues I’ve seen most often in my career.”

In workflow documentation:

“Our deploy process has these 7 steps. Here’s why each one exists and what can go wrong if you skip it.”

They’re not just using AI. They’re multiplying decades of expertise through AI.

The Real Divide

So the divide isn’t experienced vs. inexperienced. It’s not traditional vs. AI-native.

The real divide is: who’s willing to template their expertise?

Experienced engineers who do this get exponential returns. Their 15 years of knowledge doesn’t disappear — it gets amplified. Every pattern they’ve learned, every mistake they’ve made, every system they’ve built becomes encoded context that makes AI output dramatically better.

Beginners like me can punch above our weight. But we’re still limited by what we know. I’m building my expertise as I go — learning patterns, making mistakes, developing taste. AI accelerates that, but doesn’t skip it.

What Gets in the Way

If experienced engineers have such an advantage, why do so many struggle with AI tools?

A few patterns I’ve seen:

Muscle memory: When you’ve spent years getting good at typing code, your fingers itch to type. Stopping to write specs feels like overhead. It’s not — but it feels that way.

“I can do it faster myself”: Sometimes true, for small tasks. But this mindset doesn’t scale. The point isn’t speed on a single task. It’s building systems that multiply your output across everything.

Not documenting the patterns: Senior engineers often have incredible patterns — in their heads. Undocumented. If you can’t articulate it, AI can’t use it. The act of documentation is the act of creating leverage.

Treating AI like autocomplete: Autocomplete follows your lead. AI agents can lead. They can make decisions, explore options, propose solutions. But only if you give them room to.

Giving up too early: The person who tried GPT-3 in 2023 and concluded “AI can’t code” is making decisions based on ancient history. The tools have changed dramatically. Keep testing.

What Actually Works

Here’s what I’ve learned from watching engineers successfully make this shift:

Template your taste. The things you “just know” after years of experience? Write them down. What makes code good vs. bad in your domain? What are the patterns you’ve seen fail? AI can’t read your mind, but it can read your documentation.

Spec before you build. Not documentation-for-documentation’s-sake. Actual specifications: what you’re building, why, how you’ll know it works, what could go wrong. The spec is where your experience shows up.

Build verification systems, not review processes. You can’t read every line of AI output at scale. Tests, types, linting, spec-based validation. Use your experience to design the checks, then let automation run them.

Start with your highest-leverage patterns. What do you do better than anyone else? What patterns have you refined over years? Those are your first templates.

The Path Forward

If you’re an experienced developer reading this: your skills aren’t obsolete. They’re your competitive advantage.

But you have to encode them. Template your workflows. Document your patterns. Turn the expertise in your head into context AI can use.

The 15-year veteran who does this doesn’t just keep up with the AI-native newcomers. They leap ahead. Because they’re multiplying something real.

If you’re someone like me — newer to this, learned with AI from the start — keep building expertise. AI accelerates learning, but doesn’t skip it. The patterns, the taste, the systems thinking — these still matter. Maybe more than ever.

We build systems that build systems. Your engineering knowledge is what makes those systems good.