Anthropic engineer:"If your agent doesn't have the same access you do, it will never work alongside you"this 23-minute talk by a Claude Code engineer explains the exact workflows Anthropic's own team automated first:> Slack, CI/CD, internal docs - if Claude doesn't access these, it's just a terminal toy> context learning -> fine-tuningfine-tuning causes hallucinations and is too slow for fast-moving models> agentic red squiggles - soft warnings that let the model self-correct on the fly> stable