Industry Wide AI Psychosis

Summary of Main Points (AI Psychosis / Software Development Commentary)

• “AI psychosis” among companies and teams: The speaker argues that some organizations operate under an overly optimistic, almost irrational belief system about AI—leading to conversations that ignore real engineering risk.

• Misuse of reliability metrics (MTTR vs. resilience): They emphasize an infrastructure lesson: aiming for low MTTR (mean time to recovery) is not enough if it causes teams to neglect resilience. In their view, AI-driven workflows can encourage the idea that bugs are acceptable because “agents will fix them quickly,” which is dangerous.

• False comfort from “100% test coverage” and declining bug reports: A major critique is that superficial signals can hide systemic problems:

  • Test coverage can be misleading if it doesn’t actually validate real behavior.
  • Systems can look healthy locally while becoming globally complex and fragile.
  • Bug reports dropping may reflect changes in measurement or detection, not true safety.
  • Latent risk can rise while everything appears fine.

• Personal/identity attachment to AI: The speaker claims people sometimes treat AI like it’s interpersonal—so criticism of AI is perceived as criticism of the person who uses it. They describe this as causing immediate dismissals and making honest debate difficult.

• Counterpoint to extreme corporate narratives: They highlight a common “futurist” discourse from some corporate leaders—e.g., “engineering is dead,” mass layoffs, and reduced hiring for new grads—and argue these are often toxic and detached from reality. They note that customer service hiring/jobs may be increasing rather than disappearing (at least in the short term), suggesting the real picture is more nuanced.

• Why “coding is over” may be exaggerated: The speaker argues that AI can speed up parts of development (especially routine tasks), but it doesn’t eliminate the hard parts:

  • Mature codebases contain accumulated design knowledge and structure that AI can exploit.
  • In brand-new projects, the cost of wrong decisions can be much higher because the system evolves faster and at greater scale.
  • They frame engineering “wisdom” as the scarce resource—not typing/coding itself.

• Likely shift in how software is built: They speculate AI may reduce the creation of new libraries because fewer people focus on designing clean interfaces; instead, teams may move faster by bypassing deeper abstraction work.

• Uncertainty about the near future: The speaker concludes that developments are moving too fast to predict—possibly “nuts” in months—and expects rapid transformation rather than linear progress.

Presenters / Contributors

• Mitchell Hashimoto (mentioned; likely referring to Mitchell Hashimoto / HashiCorp)
• George Hotz (mentioned)
• Kramer (mentioned; associated with Century, said to be founded by him)
• Gary Tan (mentioned; example referenced)