Raw LLM Responses

Best I can say is big tech SWE's aren't driving either side of these conversations. These companies are leaning in, but the results so far are lukewarm.  Before I left my company, about the most impressive thing was AI generated updates to diffs predicated on comments, usually comments where the reviewing engineer included pseudo code or clear instructions.  Maybe the best use of AI was in Slack to provide auto responses to platform questions e.g. "is it possible to have a custom URL for a gRPC endpoint on Edge?" and it'll spit out something semi-accurate based on internal documentation and other Slack conversations of the past. My main hope is that unit testing, and eventually integration tests, can be fully automated by AI agents.  Well designed interfaces, and small constellations of these, along with well-designed IDL for RPCs, should be reasonably deterministic for AI agents to combine fuzzy LLM results with clear targets such as code coverage.