Chatbot glued to hope.
- UnderstandingPrompts + guesswork. Every model change is a rewrite.
- ExecutionAd-hoc tool calls. AI may or may not do the right thing.
- Accountability"AI did it." No trace, no audit, no request_id.
- New intentsWeeks of prompt engineering and manual QA.
- DebuggingGrep through chat logs, rerun by hand, cross fingers.
- CostPer-seat SaaS or per-token chaos. Both scale badly.
