The Programming Podcast AI That ACTUALLY Ships: JSON, Voice Agents, MCP, and Software Developer Real-World Pitfalls
Oct 3, 2025
Discover how JSON is crucial for reliable conversational AI, dramatically reducing intent errors. Learn the benefits of structured prompts in multimodal models and the importance of context windows for maintaining conversation flow. Hear about practical use cases in voice technology, like automating drive-thrus and call centers. Delve into the impact of AI on job markets, highlighting roles at risk and strategies for reskilling. Plus, get tips on negotiating job offers and applying AI practically in development.
AI Snips
Chapters
Books
Transcript
Episode notes
JSON Makes Conversational AI Deterministic
- Freeform LLM replies break downstream systems because they lack structured fields for routing, confidence, and sentiment.
- Forcing JSON outputs makes intent deterministic and enables automated routing and audits.
Enforce Fields And Confidence Floors
- Demand structured fields like department, sentiment, confidence, and content in every model response.
- Use confidence floors to auto-retry low-confidence outputs before exposing them to users.
Context Windows Drop Details; Rules Persist
- Context windows truncate earlier conversation details as token limits approach, condensing history into broader strokes.
- Persist rules in every call because rules are re-evaluated each request while context can be dropped.
