There are days when a simple thought pops into your head: “It’s just a tiny piece. An hour, maybe two.” But then a day passes, then another, and suddenly you realise you’re hunched over fixes for something that seemed trivial, your logs starting to look like a novel.
I’m getting irritated. The AI – tactfully – keeps taking breaks from the context. The tension hangs in the air as if we weren’t fixing code, but conducting peace negotiations.
And that’s exactly when you experience something on your own skin that’s not very poetic, yet painfully true: the more projects you have and the more complex they are, the faster you run into the kind of error you can’t just “push through” with sheer determination.
Even if, on the other side, you’ve got a model that can write an essay, code, a presentation outline, and still ask for water with lemon.
Back to a Method From Two Decades Ago
Looking for a way out of this dead end, I remembered Marcin Kosewski and a lesson from… well, let’s call it “more than twenty years ago.” The Kepner-Tregoe method. A classic approach to problem solving: logical, step by step, no magic.
And I felt relieved. Not because the bug disappeared. I simply stopped running in circles. This method forces order into your thinking.
Instead of “fix it, it doesn’t work,” you start asking: “where exactly doesn’t it work, when does it work and when doesn’t it – and what does that say about the cause?”
Sound a bit boring? Maybe it does. But it works surprisingly often.
A Prompt Instead of Yet Another App
Of course, my first reaction was the usual: “I’ll just write an app.” And then the second: “Bartek, don’t do it.” So I held back. At least for now.
Instead, I created something simpler: a prompt. One that, after two failed attempts to fix the same bug, switches into “stop guessing” mode. In short:
first a solid diagnosis, then comparisons in the “is / is not” style, then hypotheses and tests designed to confirm or falsify them. And when contradictions appear (“it has to be this way and at the same time it can’t be”), instead of fighting reality – you look for a workaround in the spirit of TRIZ.
That was my experiment with GPT-5.2 Thinking. And yes, you could sense the surprise. Because AI also appreciates it when a human stops saying “fix it” and starts saying “let’s check the facts.”
What Came Out of It (And Why It Matters Beyond IT)
Did the prompt solve my problem directly? No. And that’s the point: it allowed me (and the AI) to make the only reasonable decision – to back out of the dead end and look for a different way to implement one of the functions, instead of mindlessly reworking the code.
This isn’t just a programming lesson. It’s a lesson in communication, project management, life. Sometimes “pushing harder” just increases friction. Sometimes you have to step back in order to move forward.
Labs – Tools for People Who Just Want to Get the Job Done
And here comes my “by the way”: labs.trustedone.pl. It’s my experimental space where I build tools with PR and communications people in mind.
Not so that everyone has to become a prompt engineer. The point is to deliver a text, a brief, a reply or an analysis faster and with more confidence.
You’ll find apps there that do ordinary but necessary things: they check readability and content density, point out fragments where the text becomes “pretty” but loses substance, help you translate without losing meaning, and organise materials for publication. No corporate pomposity. Controlled chaos – but only in sentence structure, not in logic.
And those two new apps? They’re already sprouting in my head and I can’t wait to build them. One is meant to “break the spell” of a text so the reader understands it right away. The other is meant to shorten the path from “I have tons of material” to “I have a sensible version ready to publish.” I’ll leave the details behind the scenes – what matters is that these are tools for real work, not for showing off technology.
Beta tests are already underway. I’ve asked a few people to try them out in practice: do they work, are they non‑annoying, do they actually save time?
And I’m going back to my “simple fragment.” Only now with a different decision on the table. And with an AI that – I have a feeling – also breathed a sigh of relief.