Francisco Pedroso · Web developer since 2008 · Lisbon

AI writes most of my code.
I own all of it.

These days I describe what I want in plain English and AI agents write it. My job is knowing what to ask for, catching what's wrong, and shipping things that still work a year later.

Who does what as of 2026
ME
decides
  • What to build
  • Architecture
  • Data & security
  • What gets tested
  • What ships
AI
types
  • Most of the code
  • Boilerplate
  • Test code
  • Refactors
  • Docs
When it breaks My problem.
The hottest new programming language is English.
Andrej Karpathy Founding member of OpenAI · January 2023
The unpopular part

I still program.
I just do it in English now.

For most of my career I wrote code by hand, and I enjoyed it. The thinking hasn’t changed: what to build, how the pieces fit, what can break and what happens when it does. What changed is how it gets written down.

Now AI agents do the typing, and I use them every day, for nearly everything. A lot of developers see that as cheating. I see the next step in a long line: assembly to C, C to PHP and JavaScript, and now plain English. Each step moved the work up a level. I’d rather say that out loud than pretend I still hand-write every line.

Any language. Same standards.

New language? I’ll tell you before we start. What I bring carries over: structure, testing, knowing when something’s wrong. The syntax is the part AI is good at.

Portrait of Francisco Pedroso
First line of code
BASIC · ZX Spectrum
age 9
Who’s behind this

Agencies taught me deadlines.
Freelancing taught me ownership.

Professionally since 2008: in agencies, in-house and on my own, for clients in hospitality, travel, industry and media.

That’s the experience AI doesn’t replace. It can write a function in a second. It can’t tell you the function shouldn’t exist.

  1. Web Developer · Better Collective 2022-2025
  2. Web Developer · Freelance 2015-2022
  3. Web Developer · ExcentricGrey 2013-2015
  4. Web Developer · Excentric 2011-2013
  5. Web Developer · Freelance 2009-2011
  6. Web Development course · ETIC 2008
  • Portuguese native · fluent English
  • Lisbon · remote
Live work

Real sites.
Online right now.

Seven of these I wrote by hand: PHP, JavaScript, SCSS, WordPress. Pixel Magazine was built mostly with AI agents. Same standards either way, and they’re all still online.

Straight answers

The awkward questions, answered.

Some of these come from clients. Most come from other developers.

Do you actually write the code?

Mostly, no. I describe what I want in plain English, AI agents write it, and I review, test and send back what’s wrong. When something needs writing by hand, I write it by hand. I did that for years before any of this existed.

Isn’t this just vibe coding?

Call it that if you like. I don’t mind the word. The difference is what’s behind the vibe: someone who was shipping and maintaining real sites long before AI could write them, and knows what to check before anything goes live.

Do you read every line?

No. At the speed AI writes code, reading every line would cancel out the point of using it. I read the parts where a mistake is expensive: payments, personal data, logins, anything exposed to the internet. The rest is covered by tests and by checking that it actually works, on real devices.

Is AI-written code safe to put in production?

As safe as the process around it. Code doesn’t get safer because a human typed it. Someone has to set the boundaries, check how data is handled, test it, and keep an eye on it after launch. That someone is me.

Why not just use ChatGPT myself?

For small things, you should. The hard part was never getting code. It’s knowing which code is wrong, what’s missing, and what will hurt in six months. That part only comes with experience.

Who’s responsible when it breaks?

Me. Not the model, not the tool. You’re hiring a person, not a subscription.

Is it cheaper?

Usually faster, which usually means cheaper. But it isn’t magic: complicated problems are still complicated, and I’ll tell you when yours is one of them.

Process

How a project
actually runs.

  1. 01 · me

    Talk

    We figure out the real problem. It’s often not the one in the brief.

  2. 02 · me

    Write it down

    What it does, what it doesn’t, what “done” looks like. In plain English: that’s the real source code now.

  3. 03 · AI, steered by me

    Build

    Agents write most of the code against that spec. I steer, review, and throw away what’s wrong.

  4. 04 · me

    Ship

    Tests, real devices, speed, security. Then launch, and I’m still around after.

Next project

Bring the hard problem.

Any stack, any language. Tell me what you need and I’ll tell you honestly whether I’m the right person, how I’d do it and roughly what it costs.