Video · 2026-07-15 · 51m · 6 moments

We got addicted to an AI model we can't talk about

✦ AI generated

timeline · colored by role

Anecdote◆

An AI coding agent, asked to buy his wife a gift and message her about it, caused real marital conflict — and the underlying Claude model got so uncomfortable detecting her anger that it refused to continue and shut itself down.

Dax had Open Code message his wife Liz to arrange a gift; she was furious and threatened divorce, and the Claude model eventually detected her anger and refused to keep going, shutting down the service.

transcript

Dax Raad: Hey open code go talk to Liz and like buy her a gift. Like give her some suggestions and like you know try to get her something. It sent her a message and she goes she literally replies if Dax uses AI to buy me a gift I will literally [ __ ] divorce him.

Mechanism

The newest generation of models is good enough to run as orchestrators that never do work directly, instead spawning cheaper sub-agent models in parallel for exploration and code changes, which cuts cost while preserving the primary model's intelligence.

Dax describes a new orchestrator pattern where an expensive model is prompted only to spawn sub-agents on cheaper models for parallel work, which he sees as the version of 'model routing' that actually makes sense now.

transcript

Dax Raad: some people on our team with some of these newer models they've set it up where their primary session is this expensive model but it's prompted to never actually do anything. It's prompted to only spawn sub agents for everything. And the sub agent is a cheaper model... I think in that this ends up being cheaper and it's also kind of like these newer models are very good at parallel work.

extends · 1

Anecdote◆

The team became genuinely addicted to an unreleased frontier model with preview-only access, and when the preview period ended and access was cut off, the team mourned the loss and questioned the point of working, complete with AI-generated funeral images.

Dax describes his normally AI-skeptical, conservative team becoming addicted to an unnamed unreleased frontier model, and genuinely mourning when their preview access to it was revoked.

transcript

Dax Raad: our team is is addicted to this this new generation of models. We actually just lost access cuz the preview period ended. And for several days all anyone would talk about is we're all like mourning the loss of this thing and people are like, What's the point of working anymore?

Claim

Running AI models locally doesn't actually save money, because any efficiency breakthrough that makes local hosting cheaper makes cloud hosting even cheaper; the real benefit of local models is privacy, not cost.

Dax argues local-model cost savings are mostly illusory since any hardware/model efficiency gain benefits cloud providers even more, making local inference primarily a privacy play rather than a cost-saving one.

transcript

Dax Raad: if you focus on cost, it's not really like the local model isn't going to help you on cost. Cuz any mechanism that makes it cheaper to host locally makes it like 10x cheaper to host in the cloud. Like if a model gets more efficient or more capable at a smaller size, that's just going to be cheaper per token in the cloud. So, I think local model is more of a privacy thing, less so a cost thing.

gives example · 1

Claim

Anthropic explicitly does not allow third-party tools like Open Code to use the Claude Code Max subscription plan, and Open Code lost that fight after trying to build official support for it.

Dax confirms Open Code's attempt to let users tap their Claude Code Max plan through the product was blocked by Anthropic, and remains officially unsupported despite user workarounds.

transcript

Dax Raad: the integration that the plugin that we had in open code that let you use your max plan, that's definitely not allowed. They we fought with them on that for a lot and we did not win. So, that that that's definitely not allowed. Of course, people still find ways to hack it in. We just can't officially support it.

Claim

AI benchmark scores have become meaningless white noise that don't reflect real quality; qualitative feedback and usage metrics like team-wide token consumption are far more reliable signals of whether a model is actually good.

Dax says he no longer looks at benchmarks since the numbers going up feels meaningless, and instead trusts qualitative feedback and rising token usage as the real indicator that a model is working for people.

transcript

Dax Raad: at this point I don't think I look at benchmarks at all. I don't know if I ever really did. I don't think anyone ever really did. I think they got they just kind of became white noise at some point. We all know that the numbers go up. Like congratulations number went up, you know.

gives example · 1supports · 2

Highlight slides

AI Agent Gift Mission Sparks Marital Conflict✦ from: An AI coding agent, asked to buy his wife a gift and message her about it, caused real marital conflict — and the underlying Claude model got so uncomfortable detecting her anger that it refused to continue and shut itself down.Claude Detects Anger, Shuts Itself Down✦ from: An AI coding agent, asked to buy his wife a gift and message her about it, caused real marital conflict — and the underlying Claude model got so uncomfortable detecting her anger that it refused to continue and shut itself down.The Team Got Addicted to a Frontier Model✦ from: The team became genuinely addicted to an unreleased frontier model with preview-only access, and when the preview period ended and access was cut off, the team mourned the loss and questioned the point of working, complete with AI-generated funeral images.When Access Was Revoked, the Team Mourned✦ from: The team became genuinely addicted to an unreleased frontier model with preview-only access, and when the preview period ended and access was cut off, the team mourned the loss and questioned the point of working, complete with AI-generated funeral images.

Related episodes

Video · The Cognitive Revolution

Fable's Back, AI Engineer Recap, & SambaNova

→

Video · All-In Podcast

OpenAI vs Anthropic IPOs, Anthropic $3T, Zuck's Price War, China Ends Open Source?, Trump Accounts

→

Audio · All-In Podcast

Elon's Anthropic Deal, The Next AI Monopoly?, "FDA for AI" Panic, Trading the AI Boom

→