The Standup with ThePrimeagen Whats really going on with AI, Expert weighs in
Mar 13, 2026
Casey Muratori, veteran game developer and podcaster known for technical depth. Dimitri Spanos, AI researcher with 20+ years building production systems. They debate token costs and hardware limits. They question whether engineers will shift to reviewing AI-generated code. They unpack risks like cognitive debt, tooling vs model gains, and who benefits or loses from AI.
AI Snips
Chapters
Transcript
Episode notes
Token Cost Cuts Depend On Hidden Infrastructure
- Token cost improvements depend on hardware, infrastructure and secret algorithmic work, so 100x cheaper soon is possible but uncertain.
- Dimitri cautioned timing matters: claims (like Musk/Tesla examples) can be true but take years of engineering and infra to realize.
Treat AI Outputs As Junior-Level Work
- Don't expect full hands-off AI for complex projects; reliable hands-off generation is roughly junior-level, a few thousand lines.
- Dimitri recommends viewing current limits conservatively and using oversight, tests, and multiple attempts for larger tasks.
Early Adoption Means Lots Of Reviewing Work
- Expect a review-heavy phase where businesses push AI-generated PRs and monitor token use, creating new review workloads.
- Dimitri warns this will feel invasive and may force frequent PR reviews tied to token dashboards at some companies.