Scaling tech organizations at global unicorns like Hellofresh, Contentful and Getir since 2015. Programmer since 2001. Educator and content creator. Finding success by empowering others.

I think juniors are going to learn faster than we did because of AI.

- When I started like 25 years ago, I had to buy paper books.
- Those who started 15 years ago had Stack Overflow.
- Those who start today have a tireless, personalized tutor.

I can already see the smarter ones reaping the benefits. I know a couple.

Pedro Gil Carvalho

VP of Engineering • Sharing my journey from developer to VP • Follow for daily tips

We put my face on a video. Did you know we're hiring?

Come take a little peek into thermondo Tech - we're all here trying to figure out how to decarbonize every home, and we're using all the tech we can to get there: we build web and native apps, IoT backends, AI software, data science, pretty much the only thing we don't do in Tech is crypto.

Enjoy, and we're hiring! #teamthermondo

Find our open roles here 👉 https://lnkd.in/diZFQupk

https://lnkd.in/dd8WUdMH

Open video: Working at thermondo: Inside our Tech Team on youtube.com

Working at thermondo: Inside our Tech Team

AI-generated code is getting good enough that it takes longer to understand than to write it all myself.

One year ago, autonomous coding loops gave me broken code. It was easy to verify that it didn't work. Now they mostly give me what I ask for - but it takes real time and energy to verify.

Right now this feels like a problem because I don't trust:
1. That it didn't make a stupid mistake somewhere
2. That I specified it well enough to surface those mistakes

I don't think the solution is to slow development down to my own speed, but to improve tooling so that the kinds of checks I'm likely to want happen automatically too.

Feels like a reincarnation of a phenomenon from years ago, where I spent way more time writing tests than writing the production code. I trusted the tests - as long as they were green, I was good.

AI gives programmers hyper powers. We already had super powers.

It’s not just coding faster. It’s finding gaps in specs and designs, coming up with proposals, debating options, and generally closing loops in minutes that used to take days or weeks.

We can send AI agents on research missions and get a weeks worth of browsing distilled in 5 minutes.

Have them cross reference that against requirements, and team and tech stack preferences, and they come up with stuff that’s legitimately viable.

Even if we then code it all by hand we still can short cut days and weeks of research and debate.

Because that’s where time disappears in software projects, not so much in coding.

Does AI write bad code? Sure, it’s not perfect, but neither is the alternative.

Try getting a dozen people to agree on what to build and how and then get them to move in the same direction. 

Then at the end it’s not what anyone asked for and it’s written in a stack you didn’t expect. Half the team wanted it to be performant, the other half wanted it to be future proof, now it’s neither. But now you have to live with it because this cost you six figures in salary and overhead and the C level won’t stand for a rewrite. And it was a bit rushed so there are going to be incidents and maybe someone finds an exploit.

Go through enough of those cycles and the argument that “AI doesn’t write perfectly organized and secure code” loses a lot of its persuasiveness.

SPONSORED BY MY WIFE - She's a photographer and I was her latest victim. Best thing to happen to my linkedin in years!

I wanted to come across as "approachable but having my act together", not "just walked in from the street". And I think she nailed it.

Now every time I show up in a DM or slack message I feel like I actually belong in a professional setting.

If you're in Berlin and your headshot is quietly working against you (which it probably is), go fix it: https://lnkd.in/dchxawkU

Still thinking about juniors and the impact of AI on job prospects.

I suspect there are going to be a hell of a lot more startups and one person businesses. You don't need permission to go into business on the internet, and why would you wait for a job when you can create your own?

In 2023 I read someone saying prompt engineering was a useless skill because surely the models would get better. 

While I agree, and they have, getting familiar with how these things work still pays off. Everything is built on something that came before. Understanding the layers is an unfair advantage and you should take it.

Concrete example: Codex and Claude code have different “personalities”. You can push them down under layers of Opencode and Spec-kit… or you can learn how they respond to different inputs and get a much smoother, faster, more accurate experience. Which translates to time saved and more getting done.

taplio

Pedro Carvalho's Linkedin Analytics

Pedro Carvalho

Performance ScorePedro Carvalho's LinkedIn Performance

What is Pedro talking about?

Pedro Carvalho's Best Posts (last 30 days)

Ready to grow your LinkedIn presence?