Voice AI
Agent Validation: Stop Before You Book, Charge, or Send
Your voice agent should never book, charge, or send without clearing a validation check first. Here's the architecture that makes that happen in practice.
SB
25 June 2026The TheAutomate.io Blog
Insights on AI voice agents for Australian businesses: case studies, how-to guides, and industry trends
5 min
Voice AI
Retell's Concurrent Call Pricing: What You're Actually Paying For
Retell prices on peak concurrency, not total call volume. Twenty concurrent calls are free. After that, it's $8 per slot per month.
SB
24 June 2026
Voice AI
The Dead Letter Queue: Where Failed Calls Go to Be Replayed
When a call event fails, most stacks quietly drop it. A dead letter queue catches the failure and lets you replay it. Here's the build.
SB
23 June 2026
Voice AI
Why Production Agents Store State in Postgres, Not the Model
Storing call state inside the model's context window feels simple until a call drops or a retry fires. Here's why production agents write progress to Postgres instead.
SB
19 June 2026
Voice AI
The Quote Was $0.07 a Minute. The Bill Wasn't.
That $0.07 a minute quote sounds sharp until you see the full bill. In production, voice agent cost runs $0.13 to $0.31 per minute once the model, telephony, and TTS stack up.
SB
15 June 2026
Voice AI
Cheap Model First, Expensive Model on Retry: Voice Agent Cost Control
Running a voice agent doesn't have to mean a runaway inference bill. Here's the production routing pattern that keeps costs flat and accuracy honest.
SB
15 June 2026
Behind the Scenes
Model Dependency Killed My Build Mid-Session
On 9 June 2026 I was building on the strongest model available. By 12 June it was gone. Here's what model dependency actually costs when you're days from shipping.
SB
14 June 2026
Behind the Scenes
LLM Evals Are the Unit Tests of AI Agent Work
We started writing LLM evals after the third production regression. Turns out they're just unit tests for AI agents. Here's the build.
SB
9 June 2026
Behind the Scenes
Code Description Is Now the Bottleneck (Not the Code)
Spec-Kit reached 90,000 stars in seven months. The signal is hard to ignore: writing code isn't the problem anymore. Describing what you want built is.
SB
8 June 2026Page 1 of 5