Thursday, October 9, 2025

The Greatest Engineers Aren’t Writing Code Anymore — And That’s Precisely How It Ought to Be | by Jano le Roux | The Startup | Oct, 2025

AI didn’t kill coding. It killed the boring elements.

Press enter or click on to view picture in full measurement

Claude Code and Zed Code Editor icon.
Illustration by writer.

I’ve been coding since I used to be 12.

Began with Weebly. Then HTML. Then CSS. Then JavaScript. Then Swift. Then no matter else I wanted to construct no matter silly thought popped into my head that what

For years, I assumed being an excellent developer meant memorizing syntax, debugging for hours, and Googling Stack Overflow solutions at 2 AM.

I used to be incorrect.

The most effective engineers I do know at this time barely contact their keyboards.

They assume. They architect. They dream.

And AI does the remainder.

The dying of the code monkey

Final week, I watched a 19-year-old school dropout construct a totally practical SaaS product in 72 hours.

No workforce. No traders. No senior builders trying over his shoulder.

Simply him, his laptop computer, and Claude Sonnet 4.5.

He described what he wished. Claude wrote the code. He examined it. Claude mounted the bugs. He deployed it.

Two weeks later, he had 200 paying clients.

This is able to have taken a workforce of 5 engineers six months to construct simply three years in the past.

Now it takes one individual with a transparent imaginative and prescient and a weekend.

The foundations modified in a single day — and most of the people haven’t seen but.

Right here’s what really issues now

You don’t must memorize each JavaScript methodology anymore.

You don’t must spend 4 years getting a pc science diploma.

You don’t must grind LeetCode issues till your mind doesn’t mind anymore.

What you want is easier — and tougher.

You might want to perceive how methods work.

You might want to know what’s attainable.

You might want to clarify your imaginative and prescient clearly sufficient that an AI can construct it.

The engineers successful proper now aren’t those who can write the cleanest code. They’re those who can assume via issues, perceive structure, and talk what they need.

They’re the conductors — not the orchestra.

Its creativeness science.

The instruments that made this attainable

Claude Code modified how I work greater than any software within the final decade. It’s a command-line interface that lets Claude write, debug, and deploy code instantly out of your terminal. You describe what you need. It builds it. You evaluation it. Ship it.

I’ve been stricken by a terminal awakening.

The second I requested it to wash up a few of my previous code, I knew this can be a new stage of AI. It blew me away. What I couldn’t get proper in 100 prompts, it mounted in a single immediate. What I couldn’t get proper in 100 prompts it mounted in a single immediate. Identical to that.

It’s so out of this world f*cking good. Its so f*cking stunning to see the way it thinks.

Google CLI introduced related magic to Google’s Gemini mannequin to principally do the identical thought — AI that writes production-ready code out of your terminal.

OpenAI Codex powers GitHub Copilot and a dozen different instruments. It’s the engine behind most AI coding assistants you’ve most likely used with out realizing it.

Droid took it additional. It doesn’t simply write code — it understands your whole code doesn’t It suggests architectural enhancements. It refactors legacy legacy. It principally turns into your whole engineering workforce.

These aren’t toys. They’re manufacturing instruments that actual firms use to ship actual merchandise.

The brand new breed of code editors

Writing code in VS Code began feeling outdated six months in the past.

Not as a result of VS Code is unhealthy. As a result of one thing higher emerged.

Cursor turned my every day driver virtually instantly. It’s not only a code editor with AI bolted on to it. It’s designed from the bottom up round AI help. You may spotlight code and ask it questions. You may describe options and watch them seem. You may debug by having a dialog.

It feels much less like writing code and extra like having a technical co-founder.

Zed is my new obsession. It’s free. It’s quick. It’s stunning. And it allows you to carry your personal API keys — which suggests you’re not locked into some subscription mannequin the place you’re paying per keystroke.

I run it with Claude Sonnet 4.5, and it’s the closest factor to magic I’ve skilled in software program growth.

Press enter or click on to view picture in full measurement

Bar chart comparing AI model accuracy on the SWE-bench Verified software engineering test. Sonnet 4.5 leads with 82.0% (77.2% without parallel compute), followed by Opus 4.1 at 79.4%, Sonnet 4 at 80.2%, GPT-5 Codex at 74.5%, GPT-5 at 72.8%, and Gemini 2.5 Pro at 67.2%.
Anthropic

These editors don’t change considering. They change typing.

And that’s precisely the purpose.

The one-person startup is now a actuality

One thing fascinating is occurring in tech proper now.

Corporations that used to wish 10 engineers now makes use of one engineer.

Merchandise that took months to construct now take days.

MVPs that price $50K to develop now price $0 and a weekend.

I’m watching solo founders construct and ship merchandise that appear like they got here from well-funded startups. Stunning interfaces. Complicated backend logic. Cost processing. Consumer authentication. E-mail notifications. The works.

One individual. One imaginative and prescient. One mind.

The bottleneck isn’t technical capacity anymore. It’s creativeness.

  1. Are you able to image what you wish to construct?
  2. Are you able to describe it clearly?
  3. Are you able to assume via edge circumstances and person flows?

If sure, you’ll be able to construct it.

The hole between thought and execution collapsed.

Why understanding fundamentals nonetheless issues

Right here’s the place individuals get confused.

AI didn’t make coding data out of date. AI It made it extra precious.

The builders who perceive how databases work, how APIs join, how authentication flows, how servers scale — they’re those utilizing AI most successfully.

They know what to ask for. They know when the AI is working. They know tips on how to information it towards higher options.

Those who skip studying fundamentals battle. They’ll’t consider the code AI generates. They’ll’t debug when issues break. They’ll’t architect methods that scale.

You don’t must memorize syntax anymore.

However you completely want to grasp how software program works.

Although I hate the truth that I needed to spend hours and hours and days and sleepless nights debugging code on Stack Overflow. That I nonetheless went via that to grasp precisely how code works.

You don’t must know tips on how to combine concrete to design a constructing. However you higher perceive load-bearing partitions, basis necessities, and structural integrity.

AI is the concrete mixer. You’re nonetheless the architect.

The optimist’s view

I’ve by no means been extra enthusiastic about expertise.

Not as a result of AI is spectacular — although it’s.

As a result of it’s democratizing creation in a manner nothing else has.

The child in rural India with a laptop computer and web can now construct merchandise that compete with Silicon Valley startups.

The designer who at all times had product concepts however couldn’t code can now ship them.

The entrepreneur with area experience in some area of interest trade can construct customized software program with out hiring a dev workforce.

The barrier to entry collapsed.

And meaning the perfect concepts win — no matter who has them or the place they arrive from.

That is stunning.

My present setup

I spend most of my time in Zed today.

Claude Sonnet 4.5 handles the heavy lifting. I describe options in plain English. It writes the code. I evaluation it. Make changes. Ship it.

My workflow seems to be nothing prefer it did two years in the past.

I’m not Stack Overflow-ing syntax errors. I’m not debugging for hours. I’m not combating with bundle managers.

I’m fascinated about person expertise. About enterprise logic. About what issues want fixing and tips on how to resolve them elegantly.

The grunt work disappeared.

What’s left is the inventive work — the half that truly issues.

The long run is already right here

5 years from now, we’ll look again at 2025 because the turning level.

The second when software program growth stopped being about typing and began being about considering.

The second while you didn’t want a pc science diploma to construct world-class merchandise.

The second when the perfect thought gained — not the most important engineering workforce.

We’re dwelling via it proper now.

I really feel so extremely fortunate to be alive throughout this awakening of a brand new period. That is a lot larger than something we’ve ever seen earlier than.

  • The automotive. Folks can drives from A to B.
  • The pc. Folks can kind A to B.
  • The web. Folks can talk information from A to B.

AI means to me, should you can think about it, you’ll be able to create it.

Each morning, I get up and take into consideration what I can construct. Not what I understand how to code.

Whereas standing within the bathe, it seems like concepts roll over me like drops of water. I can not assist considering of the tune A Million Desires by Benj Pasek and Justin Paul carried out by P!nk:

I shut my eyes and I can see
A world that’s ready up for me
That I name my very own
By means of the darkish, via the door
By means of the place nobody’s been earlier than
However it seems like residence

They’ll say, they will say all of it sounds loopy
They’ll say, they will say I’ve misplaced my thoughts
I don’t care, I don’t care, so name me loopy
We will dwell in a world that we design

As a result of creativeness is the one bottleneck left.

The brand new definition of an awesome engineer

The engineers I like most at this time aren’t those writing probably the most code.

  • They’re those fixing the toughest issues.
  • They’re those considering via edge circumstances earlier than writing a line.
  • They’re those architecting methods that scale.

They’re those asking the correct questions.

Code is simply the implementation element.

AI handles implementation particulars very well.

What AI can’t do — no less than not but — is dream up what ought to exist.

That’s nonetheless on us.

And that’s precisely the way it ought to be.

So what do you do now?

For those who’ve been laying aside studying to code as a result of it appeared too onerous — begin at this time.

Not by memorizing syntax. By understanding methods.

Find out how databases work. Find out how APIs join issues. Find out how authentication retains customers protected. Find out how servers deal with visitors.

Then seize Cursor or Zed. Fireplace up Claude. And begin constructing.

Describe what you need. Let AI write it. Study from what it creates.

One of the best ways to be taught coding in 2025 isn’t via tutorials.

It’s via constructing with AI as your pair programmer.

We’re simply getting began

The instruments I described? They’ll be out of date in six months.

One thing higher is coming. One thing quicker. One thing smarter.

This isn’t the height. That is the start.

And that’s what makes it so thrilling.

We’re watching a elementary shift in how people create software program.

  • From typing to considering.
  • From implementation to creativeness.
  • From code monkeys to architects.

The most effective engineers aren’t writing code anymore.

They’re dreaming larger than ever earlier than — and AI helps them construct these goals.

I can’t wait to see what we construct subsequent.

Related Articles

LEAVE A REPLY

Please enter your comment!
Please enter your name here

Latest Articles