Its run in both self-service (output to developers), guided (output to product security oncall of security engineers) and used ad-hoc to power up manual security reviews. Depending on the accuracy of each rule and the impact of the pattern of security flaw the rule finds it is promoted to ultimately output to developers directly.
It finds about a third of the security vulns we unearth each year.
Since the GP has been deleted I'll respect that and not reference specifics, but I want to clarify for any passers by, much of the GP comment I replied to was of a detailed and technical nature, about things like performance enhancements, features and semantic analysis approaches that could make the tools useful in more use-cases - very different from the rather general and personal criticisms I see elsewhere in the nearby comment tree.
It's the suggested technical and product enhancements that I felt was potentially useful feedback, rather than any of the criticism (I can understand why those are deleted).
If your data is closer to a model where security bug hunters whose sole job is to find vulnerabilities and audit code, and it was deemed not useful in that scenario, then yes, I am at odds with your claim. Admittedly, that’s a pretty niche set of customers. If you don’t learn Semmle QL, and you aren’t writing queries, it’s probably not for you.