.NET Software Engineer & Cybersecurity Specialist | Advocate for AI on-device solutions that protect user privacy | Open University
opinions are my own
@jon_vs_moloch@JasonBotterill Good but won’t work for me, I like to own my harness.
I have been also built one for over a year, in fact I have rebuilt a few time.
Now I can use almost any mode + local model
I have my own implementation for goal command as well :)
@anshublog You’re calling developers dummies, but I think you’re missing the point. This is clearly a business/product decision. I don’t think OpenAI’s engineers had a vote on it. Criticise the strategy if you want, but blaming developers as a group is lazy.
The "Vibe-Coding" hangover is coming. 95% of the software being "vibe-coded" right now will be dead in 24 months.
Right now, we are in the peak volatility phase of the AI cycle. Bi-weekly model drops push everyone into a state of permanent FOMO. The entry barrier to building software hasn’t just been lowered, it has been vaporized.
The result? A market flooded with proof-of-concepts, "weekend wrappers", and ambitious clones of full SaaS platforms.
But there is a fundamental law of software that the current hype cycle is ignoring: Writing code is only 10% of the problem. Maintaining it is the other 90%.
I am sure the trend will shift dramatically in the next 12 to 24 months.
Every line of AI-generated code is a liability. When APIs break, underlying dependencies deprecate, or security vulnerabilities emerge, "vibe-coders" will face a harsh reality. If you didn't deeply understand the architecture when it was generated, debugging a legacy AI codebase 12 months later is a nightmare.
Industry data has shown for decades that up to 80% of a software’s total lifetime cost goes into maintenance. Right now, thousands of apps are being built without a dedicated maintainer. When the initial novelty fades, 95% of these projects will experience "software rot" because the builders will lack either the time, the interest, or the core engineering competence to refit the engine under pressure.
AI isn't a bubble, but the hype around it is. Once the curve flattens and model capabilities stabilize into predictable enterprise tools, the romantic era of "I’m rebuilding Notion in an afternoon" will end.
We aren't going back to the pre-AI era. People will always build custom micro-tools for personal use. But the macro-trend of casual builders trying to run production-grade software solo will dry up.
The industry will inevitably swing back toward a logical division of labor. Letting dedicated (smaller !) teams take the responsibility (and the stress) of long-term maintenance, uptime, and security.
Does this mean I think vibe coding has no place? No, absolutely not. But the idea that vibe coding is the future of building products is naive.
Do you disagree?
@diegocabezas01@josephmiclaus Yeah they are now following @AnthropicAI steps.
Why
Why
Seriously, why ?
Well, at least we can use our own harness, hope that doesn’t change
@LLMJunky Sorry, but I have to agree. dev tools are a dead end, ither there is a lot or OpenAI will also implement the same idea but better.
these are facts :(
@thsottiaux I think @OpenAI now want to convert codex to another vibe code thing, which the reason we are here in the first place, we do not like vibe coders.
That is for the other providers to do