Posts

The real cost of AI writing code is starting to show 👀

For the past 6 months, we’ve been coding like there’s no tomorrow: endless tokens, multiple agents running in parallel, shipping features in stacks and technologies we barely touched before 🚀

But the golden phase is fading. The signs are already here:

• Anthropic dropped their $20 subscription
• Token usage is getting tighter
• Companies are reporting hundreds of thousands to millions in monthly AI spend 💸

The pendulum might be swinging back. Smaller companies are starting to hire juniors again to write code. Larger companies are seeing senior engineers spend most of their time reviewing AI-generated output instead of building.

We’re not at equilibrium yet. That balance between humans and AI is still ahead of us ⚖️

Originally posted on LinkedIn.

How can an LLM write code? 🤔 How can it come up with ideas? 💡 How can it summarize long documents? 📄

It can feel like AI is “thinking”. But what is actually happening is closer to pattern recognition.

An LLM is like someone who has read millions of crime stories. When you start telling a new one, they can quickly guess what usually happens next. Not because they know the truth, but because they have seen similar patterns many times. 🕵️‍♂️

The same idea applies to everything else:

💻 Writing code - It has seen millions of code examples and learns common patterns

💡 Coming up with ideas - It has seen brainstorming sessions, strategy docs, and idea lists

📄 Summarizing text - It has seen many examples of long texts paired with shorter summaries

So instead of thinking from scratch, an LLM predicts what a good next response usually looks like, based on patterns learned from a huge amount of text.

In a way, LLMs are extremely powerful pattern recognition engines that learned from human knowledge at scale. 🌍

What analogy do you use to explain AI to non technical people?

Originally posted on LinkedIn.

Have you heard of the latest mental health phenomenon? Weirdly, it only affects programmers. Hence the name: Claude separation anxiety. 😰💻

Definition: when you have not seen actual code for such a long time that you forgot what it even looks like. All code now materializes through Claude. Your job is mostly nodding, copying, and saying “looks good”. 🤖✨

Symptoms:

• Mild panic when Claude says “I may be mistaken” 😬
• Inability to write a for loop without emotional support 🫠
• Reviewing code like a museum visitor observing ancient artifacts 🏛️👨‍💻

Side effects: You wake up at 3:17 AM in a cold sweat after a nightmare that the $90 subscription is now $9000. 😱

You check your email. Still $90. 🙏

You whisper thank you and go back to sleep.

Originally posted on LinkedIn.

On a busy week, I make around 10,000 code line changes a day on average. 🤯

Am I expected to review all that code? Pf. That is so last year thinking. Or maybe better said, 6 months ago thinking.

No. What I do instead is this:

• I make sure the architecture is clean, efficient, and designed for the problem I am solving. How? I ask the Agent to discuss options, pros and cons, and then I review it carefully to choose the architecture I want 🧠
• Then I ask the Agent to code while I watch an episode of Friends ☕
• Then I ask it to review the code and check for mistakes. It always finds 2–3 issues 🔍
• Then I test the application manually, and report issues back to the Agent to fix
• Then I ask the Agent to write tests covering the main use cases, happy path, and the error scenarios I discovered 🧪
• Then I tell the Agent: make the code production ready. Simple as that 🚀
• Optionally, I ask it to focus separately on error handling, security, performance, and observability
• Finally, I ask it to run build, linting, and assess the code coverage

I don’t need to review the code anymore. I review the thinking instead. 💡

Originally posted on LinkedIn.

Sam Altman once said millions of dollars are wasted on people thanking LLMs.

So people conclude you should not do it. It is pointless. Maybe even wrong.

I don’t think so.

When you talk to an LLM, psychologically you are not talking to a machine. You are talking to a mirror. Your brain still runs the “conversation with another human” mode. 🧠

And that matters.

Treat the LLM like shit and technically nothing happens. But something does change.

You.

The way you talk to an LLM is basically practice for how you talk in general. To anyone. Do it 100 times a day and your brain optimizes it. It gets cached. Autocomplete for your personality. ⚡

At some point you will slip and talk to a person the same way.

Keep being respectful and the worst thing that happens is you become a slightly better human. Horrible outcome. 😄

So yes, thanking the LLM might waste a few GPU cycles 🤖

But being thoughtful in how you write? That investment compounds.

Originally posted on LinkedIn.

Less than a year ago, a dear friend of mine (Zsolt Pocsaji) told me that very soon, software development wouldn’t need people to write code. Humans would only write code for true innovation. Everything else? Supervised AI agents doing the heavy lifting.

At first, I laughed. Internally… but still.

Then I looked at his completely serious face and realized this wasn’t another Elon-level “Cybertruck is indestructible” moment. 😄

The timeline since that conversation:

3 months: My agency rebrands as an AI-first consultancy 🤖

5 months: I’m leading AI transformation across multiple teams… with heavy resistance 🧱

8 months: I personally review <20% of the code my agent writes, while the internet is still debating “vibe coding” like it’s a philosophical movement 🧘‍♂️

12 months: I’m building software almost entirely through prompts in Claude. The industry quietly accepts the new reality. Job titles evolve from “Vibe coding cleanup specialist” to simply… “AI engineer.” 🚀

The speed of change is enormous.

For comparison: when Ryan Dahl introduced Node.js in 2009, I told colleagues that one day everything would be written in JavaScript. They laughed. Loudly. Publicly. Probably still laughing somewhere. 😅

And yet, JavaScript still needed 8 years to become the #1 programming language in the world.

So either: my friend is a genius (he is, obviously), I’m terrible at predicting the future, or… we’re living through truly unprecedented times.

Slightly exciting. Slightly scary. Mostly both.

Originally posted on LinkedIn.

Do you want to be a better vibe-coder? 😎

I have news: you need to learn to be a really bad boss. 😈

In the age of AI and Copilot, if you want to fully exploit all the new technologies, you also need to master some fundamental toxic management skills. 🤝

See, horrible bosses are great at:

Level 1:

• 🔍 Micromanaging your every action
• 🙅 Thinking you cannot do anything good enough without them
• 🔁 Double, triple, quadruple checking your work

Level 2:

• ❌ Criticise everything you do
• 🚫 Do not allow you to make any decisions
• 📈 Demand results, question the impossible
• 🎭 Manipulate you into doing something you don’t want

Ultimate level:

• 🔥 Fire you as soon as a cheaper or better model is available
• 🧠 Reset your memory from time to time (er…)

Well, it turns out LLM coding tools work best if you practice the virtues above. 🤖✨

And most importantly, do not forget to take full credit for all the hard work your LLM has done for you. 😌

No one needs to know your secret recipe. 🤫

Thank me later. You are welcome! 😄

Originally posted on LinkedIn.

I couldn’t believe it either 😅

When I first heard on Gergely Orosz’s podcast that some days Steve Yegge would change 50,000 lines of code with full vibe coding, I laughed. Impossible, right?

So I checked… and I was amazed.

Nearly 300,000 lines of code changed (added or removed) in my GitHub over the last 10 weeks. And I wasn’t even trying 😄

A few caveats:

• I was prototyping 5 different services
• One project I deleted and rewrote from scratch with a better prompt 3 times 🤷‍♂️

But still, it’s obviously not 300,000 lines of production quality code. Around 50-70% is throwaway code… and honestly, that’s exactly what I needed.

• When I don’t know what I want: I just test it out!
• When I don’t know if the solution will work at all: I test it out!
• When I don’t know if the architecture makes sense: I test it out!

No more analysis paralysis 🚀

Build, rebuild, and move in small steps toward real value.

Originally posted on LinkedIn.