I used to code in bursts. Push hard. Hit friction. Open ten tabs. Lose the thread. Reset.
This was pre-AI, around 2022 to 2023.
After a few hours, my brain felt like it was overheating. Not because the problems were hard, but because everything around the problems was.
Then I started using AI every single day.
The change wasn’t loud. It never felt like 10x productivity. It felt quieter, deeper, and far more important:
I could finally think longer without burning out.
The Real Bottleneck Was Never Coding
A typical day used to look like this:
- Start a feature
- Hit an unclear error
- Jump between docs, Slack, Stack Overflow
- Patch something together
- Forget why you did it
- Repeat
The fatigue wasn’t from building. It was from context thrashing.
Every interruption carried a hidden cost: reloading mental state, reconstructing intent, rebuilding momentum. Over time that cost compounds. You don’t notice it mid-session. You feel it at 6 p.m. when your mind is soup.
What Actually Changed
AI didn’t make me faster first. It made me lighter.
I stopped leaking energy on things that don’t deserve it:
- Rewriting the same boilerplate for the hundredth time
- Decoding cryptic errors
- Remembering API shapes I shouldn’t have to keep in my head
Now the loop feels different:
- Sketch the idea
- Let AI fill in the obvious parts
- Interrogate the output
- Refine based on real intent
I still own every decision. But I’m no longer wasting calories just to reach the starting line.
From Context Switching to Parallel Thinking
The old story was that AI lets you multitask. That’s not quite right. Humans still suck at multitasking. We just switch faster and pay the price.
What AI actually changed is this: you can keep your full attention on one layer while delegating another.
- While I’m weighing product tradeoffs, AI drafts the tests
- While I’m reasoning through architecture, AI explores concrete implementations and even surfaces the right parts of the codebase so I don’t have to hunt
- While I’m sharpening intent, AI handles the syntax
It’s less automation and more like externalized working memory.
You’re not doing more things at once. You’re holding bigger problems without dropping them.
The New Failure Mode: Losing Taste
There’s a real risk here.
If you let AI do all the thinking, you don’t get faster. You get weaker.
The bar has quietly moved. It’s no longer can you code. It’s do you know what good looks like.
My constraints are simple and non-negotiable:
- If I don’t fully understand it, I don’t ship it
- For anything on the critical path, I still start from first principles
- I use AI to pressure-test decisions, never to make them
AI is world-class at generating options. Your job is still taste, judgment, and direction. Clear intent has become the real bottleneck, and the highest-leverage skill.
What Most People Miss
The biggest shift isn’t speed. It’s energy allocation.
Before: Energy went into syntax, debugging, recall.
Now: Energy goes into architecture, product sense, tradeoffs, and systems thinking.
That’s an entirely different job. And it compounds in a completely different way over time.
How to Start Without Overthinking It
Ignore the hype cycles. Start small and stay honest with yourself.
- Use AI to explain errors
- Use it for glue code and scaffolding, not core logic
- Use it to generate options when you’re stuck
- Always review with intent
Then track one simple metric: How you feel after a long session.
Less drained means you’re using it right. More detached or fried means you’re leaning too hard.
The Quiet Part Everyone Underestimates
This isn’t a loud revolution with a single before and after moment.
It’s gradual:
- Fewer breaks you didn’t choose
- Longer stretches of uninterrupted thought
- Less end-of-day exhaustion
You close your laptop and realize: you did more meaningful work, and you still have energy left for the rest of your life.
Where This Is Going
The next shift isn’t coming. It’s already landing.
Today’s tools have moved past shallow file-by-file help. They now give real repo-wide awareness: seeing how services connect, what dependencies actually matter, what breaks if you change one piece. The best ones act like persistent architectural memory, holding context across days, not just the open tab.
We’re heading into agentic territory: AI that doesn’t just write code but plans, tests, iterates, and coordinates with you like a true collaborator. The tactical work (implementation, boilerplate, debugging) is shifting to the machines. Engineers are becoming orchestrators, focusing on system design, tradeoffs, and judgment calls.
Domain and architecture knowledge has only become more valuable. The leverage is real, but it is bounded by how well you understand your own system. That is exactly what I am seeing in my own flow right now.
Final Thought
AI didn’t make coding easier. It made it sustainable.
And in a world that keeps speeding up, sustainability is the bigger deal.