github.com
Blog post here: https://www.kaelio.com/blog/open-source-anthropic-internal-d...
Would love your feedback.
> When an agent writes code, the solution space is open-ended and tests catch hallucinations. Analytics is the opposite. "There's often only a single correct answer using a single correct source,"
As someone who is data-native and software-second, I would say the opposite is true. It's very difficult to specify desired code behavior in natural language, but good PMs occasionally manage it.
Beyond really trivial questions, it's much, much more difficult to specify an analytics question in natural language unless you are a statistics expert. The garden of forking paths is immense.