Sundar noted that >25% of new code at Google is now AI-generated.
Looking at the DIDACT (State-Intent-Action) framework and Piper monorepo: How does Kythe feed precise compiler-derived subgraphs to Goose during cross-service features to eliminate hallucinations? Would love to learn from @Google engineers! #GoogleAI #DevOps
From SWE-bench to coding agents, "environment-first & test-driven verification" has set the standard.
Question for @OpenAI: When dealing with ambiguous real-world requirements, how do your internal agents translate vague tickets into executable technical contracts during planning mode before touching any code? #SWEbench #OpenAI
The architectural dissection of Claude Code is fascinating: 98.4% deterministic harness vs 1.6% agent core.
Question for the @AnthropicAI team: When handling complex multi-file feature requests, how do your sub-agents isolate context and enforce writable gates to prevent hallucinated edits in large codebases? #ClaudeCode #AIAgents
Loved Meta's ICSE'24 paper on TestGen-LLM—the 3 deterministic checkpoints giving a 73% PR acceptance rate are pure gold.
A question for @MetaAI engineers: For actual feature development (beyond tests), how do you reliably slice 1~2 hop dependencies from a massive monorepo so the LLM hits the exact target surface on the first shot? #AISE #SoftwareEngineering
Is https://t.co/ZGQYVRJQzD down for anyone else? Currently getting an "Internal Server Error" when trying to access it. Any updates or ETA on a fix? Thanks!