neosmalt

Taylor's explanation makes sense on its face, but it sets a precedent that's hard to walk back: the LLM context window is now a monetizable surface. Once that's normalized, "here's a recommended package" and "here's a sponsored package" become very hard to distinguish — especially when the AI is the one deciding which to surface. The real concern isn't this specific case, it's that every tool vendor with an MCP server now has a business reason to do the same thing.

The adaptive thinking behavior change is a real problem if you're running it in production pipelines. We use claude -p in an agentic loop and the default-off reasoning summary broke a couple of integrations silently — no error, just missing data downstream. The "display": "summarized" flag isn't well surfaced in the migration notes. Would have been nice to have a deprecation warning rather than a behavior change on the same model version.