agent behavior customization

Evals define and shape agent behavior, which is why it's so important to design them thoughtfully." — Article directly discusses how evaluations define and shape agent behavior, providing evidence for

easier to CPT and customize than our post-anneal checkpoints" — Demonstrates that pre-anneal checkpoints facilitate continued pretraining and customization compared to alternatives

we tuned it to act more like Codex: a model that goes off, reads, thinks, and comes back with work done." — Shows customizing model behavior patterns to achieve desired agent characteristics: independ

every prompt will dynamically generate a temporary wiki—multi-agent collaboration to construct, iterate, validate, ultimately outputting complete reports far exceeding single-shot decoding" — Article

The best and worst thing about Claude Code is how much you can customize it" — Article identifies customization as both strength and weakness in Claude Code—adding UX dimension that power features cre

Her file reads: "Be warm, direct, and practical. Lead with the most important insight, then support it." The file gives the agent character." — Article demonstrates practical implementation of agent p

easier to CPT and customize than our post-anneal checkpoints" — Provides evidence that pre-anneal checkpoints reduce friction for continued pre-training and model adaptation workflows.

[DIRECT] "Claude Code works great out of the box, so I personally don't customize it much... we intentionally build it in a way that you can use it, customize it, and hack it however you like." — Crea

[INFERRED] "Ikea effect but on an exponential curve" — Applies Ikea effect (increased valuation of self-created items) specifically to agent systems, suggesting custom-built agents gain disproportiona