Posts

I asked my agent to find my Windows PC on the home network. It tried and concluded it was not possible.

I was tired, and for some reason I told it:

“It feels like you are not even trying.”

It was supposed to be a joke. ChatGPT would laugh and send me back a pun. 😄

Not Claude. Claude is like an autistic overachiever. “Fair challenge,” and it tried again. This time, it came back with the right IP. 🤖

It also killed my Wi-Fi, and I spent 2 hours debugging my home network, only to find that restarting my laptop would make everything go back online again. 😅

Agents are a double-edged sword. ⚔️

I asked it to debug an issue, and I spent 30 minutes convincing it that the issue was real and that I didn’t hallucinate it.

Another time, it didn’t ask back. It just fixed an imaginary issue that was the result of me phrasing the question wrong.

So, if we want to use agents deterministically, i.e. force them to produce the same results every time, then we need to apply a lot of engineering around the agent to make it happen:

• custom-written tools
• custom-written harness
• context engineering
• specific skills
• subagents
• custom workflows

Or, we can just live with the non-determinism and use trial and error. 🎲

A big part of successful communication at work is how you handle disagreements. Most people know how to do it… ineffectively. 😅

First, you listen, then you start your response by acknowledging the other party, and then you say… BUT.

“Yes, X can work, but I think Y is better.”

And instantly, the dynamic is changed forever. 💥 This is because deep down, everyone knows that everything you say BEFORE the BUT is bullshit. That simple little word instantly changes how the other person will hear your message: they won’t. They instantly know that you are invalidating them while seemingly acknowledging their point of view.

The person is instantly on the defensive. 🛡️ They stop listening and start preparing to defend their point of view. And they cannot wait for their turn to go back and argue more for their original point. This is the perfect recipe for never-ending arguments and people ghosting each other. 👻 Not the outcome you hoped for when trying to get people to do things.

So, here is the solution: replace the BUT with an AND. 💡

“Yes, X can work, and I think Y is even better.”

Instantly feels different. Saying AND means that you extend what was said before you. You add to the conversation; you don’t contradict or dismiss it. 🤝 The person feels you are on their side, and together you are even better. They will not feel the need to defend themselves, since AND means you are both right, and together you can work out what is best.

Saying BUT means you may just want to get your way; saying AND invites collaboration. 🤝 Saying BUT shows that you put yourself above the other person; saying AND puts you next to them. Saying BUT uncovers your negativity, contempt, or criticism; saying AND prompts a connection on positive grounds. 🌱

Try it yourself. Replace the BUT with AND, and see the difference in the reaction. Then share your experience. 👇 Funnily enough, this is a lot harder than you might think. 😄

I first heard of this technique from George Abonyi, and I needed 8 more years to be confident enough to write this post. Thanks, Gyuri! 🙏

You can assess an engineer’s agentic adoption by asking a very simple question: Do you still write code manually? 👨‍💻

If the answer is a straight no: good adoption. ✅ Anything else: no adoption.

A person needs to let go of writing code manually and learn how to make everything safe, reliable, and much faster agentically. 🤖 The switch is almost like a toggle, but one that takes a bit of a leap of faith.

So far, I’ve encountered the following concerns when talking about making that leap:

• 😨 Am I going to lose my job? Are all of us?
• 🎛️ How do I stay in control?
• 👀 How can I review all the code the agent writes?
• 🔐 Is this safe?
• 💥 This company just went bankrupt because of AI.

Talking through all these concerns openly is a good start, and leading by example helps people make the leap. 🚀

Hey Claude, optimize all my iOS apps for the new iPhone Duo. Make no mistakes. 😅

🍎 iPhone Duo: https://lnkd.in/dzapxmX3

Welcome to software engineering in 2026.

💡 Fun fact: add “agentic engineering” to your profile, and suddenly you’re sought after for AI Engineer roles.

But the whole concept of an “AI Engineer” seems wrong to me. 🤔 Suppose your organisation has an AI Engineer. What will all the other engineers work on? The non-AI stuff?

The reality is that all engineers are becoming AI engineers. And they need to.

“Oh, but our AI Engineer will build chat interfaces and agents.” 🤖 Okay. How is that fundamentally different from building loops, workflows, and API integrations? It sounds suspiciously like… software engineering. 👨‍💻

You don’t need one designated AI Engineer tucked away in a corner. You need agentic engineering coaches and a transformation of your entire engineering organisation. 🚀

If your AI strategy is simply “hire an AI Engineer,” you’re still doing AI the lazy way.

Software engineering in 2026: 🤧 You sneeze into a microphone 🎙️, and working software comes out the other end. 💻✨

Change my mind. 😎

I am still reading a lot of posts on LinkedIn that pose questions such as:

a) Would you switch to agents writing code? b) Can agents write good code? c) Is it economical to have agents write code?

My dudes, I have bad news for you: you are stuck on the wrong questions.

The reality is this: everyone uses agents now because everyone sees their superiority. There is no point in arguing about IF agents will take over. They already did. You just haven’t checked your mailbox yet for the invite.

So the real question is this: do you go agentic, or do you get left behind?

There is no other option for businesses. If you fail to act now, someone will vibe-code your product tomorrow, and soon you’ll be out of business.

Want to get on board faster? Gimme a call, baby!

AI will increase demand for software and engineers rather than replace them. The early signs are starting to show. 🚀

Both Microsoft and Amazon recently announced a combined $3.5 billion investment in new organisations whose sole purpose will be to work with customers and integrate AI technology into their businesses. 🤖

Every single engagement that these engineers work on will then be an opportunity for customers to build internal engineering teams and replace expensive external providers to maintain and replicate the know-how in new initiatives. 👩‍💻👨‍💻

Eventually, demand will grow as companies see the proven value in their businesses. This is demonstrated in a recent analysis from Ramp that covers 21,000 US firms: “Firms that adopt AI grow headcount by 10.2% over the two years following adoption.”

The future is bright for software and engineers. ☀️