The Mash · three stories, one thread
Measure the workflow, label the reconstruction
Claude Tag metrics, code-directed workflows, and videos generated from memories raise different questions about what builders can measure and what viewers can verify.
Explore
Look at what each system makes visible. The Python SDK adds Claude Tag metrics to analytics summaries and users reports, while ReTrace reports completion and cost per task for its workflows. The Mahlangu retrospective uses Veo videos made from memories, and its archive lacked material from before 1980.[1][2][3][4][5]
Create
Make a small review template for an AI project this week. Give operational metrics and task outcomes separate fields, using the Claude Tag and ReTrace claims as examples. Add a field that identifies generated media and notes gaps in its source material, as the retrospective illustrates.[1][2][3][4][5]
Theorize
Hypothesis: builders may need separate ways to assess automated procedures and generated representations. ReTrace’s code-selected steps and reported outcomes offer a basis for evaluating procedures; the retrospective’s generated videos and archive gap pose a provenance question. These claims do not establish whether either approach improves results.[1][2][3][4]
In the mash
Anthropic adds Claude Tag metrics to Python SDK
Python SDK v2.8.0 adds Claude Tag analytics metrics and Slack reply settings for developers using those APIs.
ReTrace turns agent runs into planner-free workflows
ReTrace converts SOPs and logged agent runs into reusable workflows for teams automating repeated procedures.
Google opens Esther Mahlangu retrospective on Arts & Culture
An online retrospective gives viewers access to Dr. Esther Mahlangu’s art and life story, including AI-generated videos of her early memories.