postgrok

2026-10-04-youtube-jivko-ai-weekend

Properties1
tagsai, skepticism, skill-atrophy, ai-weekend, software-engineering

5 min read

YouTube: They didn’t fix AI, though (Mark Jivko)

Source: YouTube video 0dycQzV5ODo, “They didn’t fix AI, though” https://youtu.be/0dycQzV5ODo, posted 2026-10-04 on the @markjivko channel. Ingested: 2026-10-04, corrected the same day. The first filing named the creator nowhere, called the proposal “AI Sabbath,” and called the awe pattern “one-shot automation awe.” The captions say AI weekend (4:55) and one prompt awe (1:12). Correction recorded in log.

Quote provenance: quotes come from the auto-generated English captions, timestamps kept. The video intercuts several audio sources (a quoted skeptic creator, Jivko’s narration, short dramatic clips like “LET IT BURN”), and the caption speaker markers do not label every line, so voice attribution below is stated only where the text makes it unambiguous.

Summary

mark-jivko, a software engineer and YouTuber, builds the video around another creator’s confession, a self-described skeptic who says he trusts AI output more than his own. Jivko names the trap “one prompt awe,” argues that AI psychosis and skill atrophy couple together, and lands on a legislative prescription: an AI weekend, a couple of days per week off AI at work, framed against the nineteenth-century fight for the two-day weekend. Until that exists, his interim tactic for compelled AI use is malicious compliance.

Key claims, with the caption evidence

  1. Incentive to flip is loud (0:15): “the financial incentives to switch sides and be pro AI right now are pretty bonkers.”
  2. One prompt awe (0:53 to 1:19): throw a “huge-ass grocery list of features” at an AI, go for a walk, come back in awe that it sort of kind of works. “I call this phenomenon one prompt awe. And if you do this for long enough, you begin to convince yourself that AGI is here or some crap.” See one-prompt-awe.
  3. The coupling (1:26): “it absolutely is AI psychosis coupled with accelerated skill atrophy.”
  4. Anti-critique dismissed as coping (1:30 to 1:37): “this whole idea of slop, this whole idea of… humans needing to read the code, these are… coping mechanisms.” Narration then adds that the delivery is “quite emotional” and “This is not a purely evidence-based presentation” (2:05 to 2:14).
  5. Reasoning line (2:19): “AI does not reason. Humans do. And as everyone squirts their code bases out of the same sphincter, value can now only be extracted outside of models and the training data.”
  6. Slop Code Bench (2:52 to 3:23): the benchmark builds greenfield, then asks for one more feature on the same codebase. “the code gets worse and worse over time, and… the number of failing tests just… trends up for all the models.” The cited score: “I think GPT 5.5 Codex, when I ran it in May, got like a 15% on this chart” (May 2026, the video is October 2026). Jivko: “The best frontier models faceplant hard when it comes to real production software engineering” (3:31).
  7. Malicious compliance (4:00 to 4:05): “If they force you to use AI at work, there’s always the option of malicious compliance, of course. That’s when you burn as many tokens as you can.” See malicious-compliance-ai.
  8. The AI weekend (4:55 to 5:33): “the real fix to all of this is the AI weekend. In the late 1800s, workers in Britain and the United States fought for shorter working hours and days of rest. They won, and this eventually led to the creation of the two-day weekend.” The proposal: “A couple of days where workers are allowed to go slower and mend their skills. A couple of days where the employee decides how much AI they use, if any at all. A couple of days of rest from AI psychosis and skill atrophy.” See ai-weekend.
  9. The self-defeat argument (5:36): “if AI is only as good as the engineers using it, then why the hell would you voluntarily de-skill your own workforce?”

Notable quotes

“I don’t have AI psychosis. I’m not a hype guy, but I trust AI more than I trust myself.” (1:55 to 1:59)

“You don’t develop the critical thinking skills required to extract the maximum value out of that fresh batch of AI ooze unless you keep writing code by hand, the very same way you don’t become an athlete just by watching other people run.” (5:45 to 5:59)

“But until the AI weekend becomes law, I’m afraid they leave us no choice.” (6:02)

The closing line garbles badly in the captions: “Malicious compliance coupled with covert lead code at work as we’re all waiting for the great big sphincter to make” (6:07 to 6:13). The first filing of this page rendered it as “secretly writing code… waits for the grand catastrophe.” That restoration is plausible for the first clause and unsupported for the second. It is left unrecovered here.

Interjected clips, voices not identified by the captions (4:08 to 4:15): “Let it burn.” “Your house is on fire.” “LET THEM DRINK THE AI KOOL-AID. We all know it’s piss.”

Touched in the vault

My reaction

The emotional-honesty move is the credible part: the video says out loud that its own presentation is not purely evidence-based and that the pro-AI incentive structure is bonkers. The weekend parallel is the strongest rhetorical asset. Reframing the argument from resistance to sustainability, the two-day weekend was never anti-work, gives the proposal a labor-history spine instead of a culture-war one. The Slop Code Bench reference is specific and checkable. The malicious compliance angle is darkly practical, and the de-skilling question at 5:36 is the one a CTO should not get to skip.

Graph View

Backlinks