Datasette-llm removes model configuration friction for agents

The actual win here is architectural: by reading the parent LLM CLI's model config, datasette-llm eliminates a second configuration step that would otherwise force users to specify their model choice twice. This is a small change with outsized impact on usability.

This mirrors the infrastructure-first thinking Anthropic outlined in their agent systems work (late October). Where Anthropic isolated execution environments to prevent cascading failures, Willison is eliminating configuration cascades that create friction. Both approaches recognize that agent reliability depends on removing decision points where users can misconfigure. The Granite Switch work from IBM also reflects this pattern: treating model deployment as a continuous, low-friction process rather than a discrete setup event. For data-centric agents specifically, reducing setup overhead directly affects whether teams actually deploy them.

If Datasette Agent reaches 0.1 stable release within the next two months with zero breaking changes to the LLM CLI integration, that signals the configuration inheritance pattern is stable enough for production use. If instead there are revisions to how parent config is inherited, that indicates the abstraction still has rough edges.