AI That ACTUALLY Ships: JSON, Voice Agents, MCP, and Software Developer Real-World Pitfalls

What do JSON and conversational AI have in common? They are the glue behind ordering coffee, booking flights, and talking to support. In our tests, about 1 out of 3 replies missed the intent until we enforced structured JSON outputs. In this episode, Danny Thompson and Leon Noel break down how to move from “cool demo” to production systems that route, escalate, and self-audit reliably.

SITE https://www.programmingpodcast.com/

💡 Sponsor: Level Up Financial Planning

Changing careers or increasing your income? Get financial clarity with Level Up Financial Planning—helping early and mid-career tech professionals secure their financial future. Visit LevelUpFinancialPlanning.com for a free consultation!

https://www.levelupfinancialplanning.com/

Stay in Touch:

📧 Have ideas or questions for the show? Or are you a business that wants to talk business?

Email us at [e-mail]!

Danny Thompson

https://x.com/DThompsonDev

https://www.linkedin.com/in/DThompsonDev

www.DThompsonDev.com

Leon Noel

https://x.com/leonnoel

https://www.linkedin.com/in/leonnoel/

https://100devs.org/

📧 Have ideas or questions for the show? Or are you a business that wants to talk business?

Email us at [e-mail]!

What you’ll learn

- Why freeform paragraphs fail backends and how JSON fields fix routing

- A simple schema pattern: department, sentiment, confidence, reply

- Confidence floors that trigger automatic retries before users ever see a response

- Context windows: why rules are read every call while context gets dropped

- MCP basics and how domain context avoids bad translations and metaphors

- Where voice agents work today (predictable conversations) and where they do not

- Practical tool choices for text, code, and voice workflows

- Real labor impacts, retention insights, and reskill advice

- Salary negotiation quick hits: the two lines that matter

Chapters

00:00 Cold open: JSON as the glue + the 1-in-3 miss

00:30 Show intro and promise

01:10 Quick definitions: JSON, NLG, NLU, MCP

03:00 Why structured JSON beats paragraphs

07:30 Confidence scores and auto-retries

10:30 Sponsor break

11:30 Prompts for image and video models that actually work

15:00 Context windows and durable rules

20:00 MCP in practice: local dialects and domain knowledge

26:00 Voice agents: predictable vs unpredictable conversations

33:00 Jobs, retention, and reskilling

40:00 Question of the Day: salary wiggle room

46:00 The Developers Guide To AI