Answered from Brian Kernighan: Inside Bell Labs, The Best Programmer, Why C Won · Ryan Peterman · 2026-10-26
Brian Kernighan said he uses mostly Claude to turn natural-language questions into SQL for a non-programmer colleague's database at the Institute for Advanced Study. About 90% of the time the query is fine and 10% of the time it is wrong, and he cannot tell which, because he is not an SQL expert. He concluded that domain expertise and ways of validating output are still required, and he doubts the models will ever be perfect. He compared the shift to Fortran in the 1950s, when John Backus worried programmers would reject compiled code as less efficient than assembly.
“And my observation is 90% of the time it generates a perfectly fine query” Brian Kernighan · at 1:14:27 — open in the transcript →
Source: our own transcription of the episode. The full conversation is on Ryan Peterman (2026-10-26). Read the whole episode summary →