Claude's UV sky map shows what backlog work AI can clear

Astronomers have complete sky maps from radio to gamma rays, but large stretches of the sky have never been observed in ultraviolet light. According to Anthropic, astrophysicist Brice Ménard guided Claude Science to find existing datasets, combine them, and fill the missing regions with statistical inference. Anthropic says the job would have taken humans weeks. With Claude it took a few days, while Ménard worked on other projects.

The category of work interests me more than the map. Anthropic frames it as slow, low-priority work that scientists never get around to, and every organization has a pile of that: datasets that should have been merged years ago, reference tables nobody owns. Finding sources, reconciling them, and estimating what is missing is data engineering. An agent doing it under an expert's supervision is a more convincing use of these models than another chat window. Still, this is one case study from the company that is selling the tool, so take it with a grain of salt.

Parts of this map come from statistical inference, not from observation. That is fine for a teaching tool, which is how Anthropic describes it. Anyone reusing it for research needs to know which regions were measured, which were estimated, and how confident each estimate is. The same holds when this pattern moves into business data. An agent that fills gaps should return a record of every value it inferred, or the guesses become indistinguishable from the measurements.