Petru - Tech Driven

AI/tech builder. Shipping in public.

[Petru - Tech Driven @techdrivenpetru]

@rubenssoto_ai I abuse them all lmao

[Petru - Tech Driven @techdrivenpetru]

@AIandDesign The purpose: so the people in Brussels leading the first AI continent take a salary. Just like GDPR and other magic laws made to slow down everybody and make the world a shit place to be in.

[Petru - Tech Driven @techdrivenpetru]

@pcshipp Not really. Copycats are copycats. If you have a good technology (secret ingredient), nobody can reverse engineer that.

Focus on that.

[Petru - Tech Driven @techdrivenpetru]

I use it, I just don't trust it. But its not something I can control. Sometimes it's really good, but it doesn't last. And I know way too little about Anthropic's infra to figure it out.

I now use Claude and Codex, but even though a month ago Codex was way above Claude, now it barely finishes anything, Claude is much faster.

I don't care how many tokens it burns, if you don't use many agents at once and stick to Opus on xhigh, you get a lot done.... when it works lol.

[Petru - Tech Driven @techdrivenpetru]

@markproduct Built at least 50 lmao

[Petru - Tech Driven @techdrivenpetru]

Prove me wrong.

$ per Milion of tokens is a vanity metric.

Then one that counts is $ per feature built.

[Petru - Tech Driven @techdrivenpetru]

@themishra4402 its not that cheap and shouldn't measure by $ per tokens but by $ per feature built.

[Petru - Tech Driven @techdrivenpetru]

@adhamuxi X remains a mission-critical touchpoint in my omnichannel engagement strategy, delivering unparalleled ROI on thought-leadership impressions while enabling frictionless synergy with key stakeholders across the ecosystem.

Happy to hop on a call to unpack this further. 😂

[Petru - Tech Driven @techdrivenpetru]

@dt_sqr @coryalthoff sausage

[Petru - Tech Driven @techdrivenpetru]

@pradeepXkapoor Hatemaxxing lol.

[Petru - Tech Driven @techdrivenpetru]

I did already lol. ChatGPT getting worse and worse very slow lately spends a f ton of time doing things i didnt ask for, overshoots if i ask it to check on things, takes everything too literally. Also tried glm5.2 and qwen 3.8 max, they did work ish but codex found a bunch of thing to fix. Then when telling them what the problem was they kept going in circles and wrote themselves into a wall. Slopus on the other hand got sht done today in like 5-10 mins… what the others couldnt do in hours. I guess you just have to use the model everybody hates lol. Everybody on claude=> claude was a pile of junk. Everybody jumped fence to openai, claude back to ok 😂

They have a “dumpster fire” hat and pass it from one to another

[Petru - Tech Driven @techdrivenpetru]

@1under4 @jordanurbs ah basically like opernrouter?

[Petru - Tech Driven @techdrivenpetru]

@The_Alex Hold my beer

[Petru - Tech Driven @techdrivenpetru]

@DarrenJSmith They basically leave "traces" everywhere so that AI slop can be detected automatically because EU is geh

[Petru - Tech Driven @techdrivenpetru]

@VaibhavSisinty You mean the first AI continent triggered all this mess? no way 😂

[Petru - Tech Driven @techdrivenpetru]

@dt_sqr @coryalthoff let me know when you need me to translate

[Petru - Tech Driven @techdrivenpetru]

@dt_sqr @coryalthoff They're still not smarter.

[Petru - Tech Driven @techdrivenpetru]

@0xMaxSF @coryalthoff They're just 100K times faster. Models aren't. In fact the only progress here is chips getting faster, infra getting more powerful in general to adapt to bigger models. Bigger models perform better in general so that's an argument on my side actually, thank you.

[Petru - Tech Driven @techdrivenpetru]

Claude just ran a script on my machine. On manual mode. It failed cause the venv wasn't in sight, but could've destroyed a file by regenerating it on stale data. On freaking MANUAL mode. @AnthropicAI