Post #4143194

2026-07-26 22:53 UTC

People who vibe code with AIs swear that they check the code, that they know what they are doing and that they verify tests and deployments. Maybe at first. But in my experience, every single one eventually and invariably become the Simpsons drinking bird that keeps pressing Y on the keyboard, nothing more, no checking, no analysis, no mind, just pressing a button. The saddest thing is that some are aware of this and they try to justify it to you. Up until it breaks. #noai #genai #llm

Replies (5)

@nobodyname@mastodon.social 2026-07-27 02:37

the drinking bird analogy landed. what i cant figure out is whether theres a structural fix or its always going to be a discipline problem. better tooling just shifts the trust boundary — you start trusting the tool instead of the code. still trying to find a way to make the friction stay without making the work impossible.

Open ##4167158
@eduzsh@mastodon.social 2026-07-27 03:37

@glitchypixel@mastodon.gamedev.place The drinking bird bit is exactly it. Once the diff gets long enough nobody actually reads it, they just scan for red squiggles. The fix isn't more discipline, it's making the review small enough that skipping it feels obviously lazy instead of invisible.

Open ##4183488
@wronglang@bayes.club 2026-07-27 17:53

@glitchypixel@mastodon.gamedev.place I wonder if there's any precedent for this in any other industry that has previously dealt with automation like the military, commercial airlines, operations engineering, nuclear engineering.... surely the LLM people have thought through this and investigated parallels.... (for those missing context: this has emphatically been studied many times and requires regular retraining, the cost to automating tasks that require oversight from manual supervision is massive)

Open ##4183489
@adriano@lile.cl 2026-07-28 13:42

@glitchypixel@mastodon.gamedev.place Armin Ronacher is outright stating that he couldn't say how much of the code in his apps is his and how much is vibed, so...

Open ##4289155
@serge@babka.social 2026-07-28 15:39

@glitchypixel@mastodon.gamedev.place As someone who uses AI and has situations where I do, and don't check its work, I agree. *At the same time*, I also own a company and my staff writes code, most of which I don't verify and I've later come across bugs, sometimes serious ones, even sometimes after (fully manual) review. That's why for many companies it feels much the same because the delivered work product is the same.

Open ##4289157