What we’re reading…

The hard parts of an agent that acts.

We build an agent that makes phone calls for you. These are our notes on what’s actually hard about that, grounded in the research.

June 2026

Tell Them It’s AI

An AI that answers your calls or books for you raises one question the law keeps answering the same way: does the person on the other end have a right to know it’s not human? California says yes. The FCC says yes. The EU says yes. And the disclosure they require is becoming a floor the system can’t switch off.

Read →
June 2026

Hacking the IVR

Press 1 for billing. Getting an agent through a phone tree is one of the hardest things you can ask it to do, because the IVR breaks the assumption most agent systems are built on: that you can try again.

Read →
June 2026

Looking Is Already an Act: The Commit Problem in Voice Agents

To talk at human speed, an agent has to start working before it knows exactly what’s needed. Fine when the work is a database read. A trap when it’s placing a call, sending a text, or canceling a booking. What makes an agent trustworthy isn’t what it does. It’s what it won’t do on a guess.

Read →
June 2026

All Talk, Some Action

Speech-to-speech models are now good enough that the voice is rarely the first thing that fails. Then you ask one to actually do something — check a calendar, call a supplier back, sit through an IVR — and the seams show. The unsolved frontier in voice agents is acting on the world without breaking the conversation.

Read →