Field notes on building voice & chat agents.
Honest writeups on latency, evaluation, prompt design, and the surprisingly squishy parts of putting AI in front of real customers.
What makes an AI voice agent feel real-time, and why it matters
Why response speed makes or breaks a voice agent on a phone call, and how Hania keeps conversations fast, interruptible, and natural.
Transferring a live call to a human, the right way
When a caller wants a person, the agent hands the live call off cleanly, with no loops and no starting over.
One agent, every channel: phone, web, and SMS
Build one agent and put it on the phone, your website, and SMS, with the same knowledge and abilities.
Build a tool by describing it: how Describe-with-AI works
Build a custom tool by describing what it should do, or pasting an API doc link, and test it before it goes live.
Why an AI agent should read the result before it acts
What separates a real agent from a chatbot: it reads each result and decides the next step.
Memory that respects scope: per-user, per-bot, per-workspace
Agents that remember customers across conversations, scoped so details stay where they belong.
Keeping personal data out of your logs
The safest place for sensitive data is nowhere. How Hania redacts personal info in real time.