The Mash · three stories, one thread
Three releases, three points in an AI build workflow
The shipped changes cover an SDK endpoint, a local model, and agent evaluation instructions.
Explore
Look at how Claude Tag endpoints fit into a Claude Enterprise organization’s Python SDK use. Separately, test whether Nemotron 3.5 Bolt E4B’s 128K-token context window and DFlash draft matter for your local workload. Check how CrewAI’s revised AGENTS.md instructions affect login during evaluation.[1][2][3][4][5]
Create
Make a small evaluation checklist this week with separate tracks for Claude Tag endpoints, the Ollama model, and CrewAI login behavior. Record whether the CrewAI evaluation starts login and stops after three rounds. Keep the tracks separate; the claims do not establish that these products work together.[1][2][3]
Theorize
Hypothesis: builders may need more explicit evaluation steps as SDK endpoints, local models, and agent instructions change independently. These releases provide three examples to watch, but they do not establish a broader trend.[1][2][3]
In the mash
Anthropic adds Claude Tag endpoints to Python SDK
Version 2.3.0 adds Claude Tag endpoints for Claude Enterprise organizations to Anthropic’s Python SDK.
Ollama adds NVIDIA Nemotron 3.5 Bolt E4B
Ollama v0.40.10 adds a local vision-language model for people working with images and tools.
CrewAI releases 1.15.30 with AGENTS.md fix
CrewAI 1.15.30 changes AGENTS.md login instructions during evaluation and updates documentation for users of the release.