Notes from The Human AI Company

A face and voice that holds a live, unscripted conversation.

Why agents need a face and a voice, not just a chat window.

Live conversation, not a chatbot and not a recorded clip.

Where a live agent fits the funnel, and how it beats an SDR.

When live beats a rendered clip, and how to build one.

A free community space and rooftop for AI builders in Berlin.

Ship a live agent without a sales call.

Why the region it runs in is a compliance decision.

Face and voice under 200ms, and the latency budget behind it.

A still face and a moving face are held to different standards. What still triggers the uncanny valley in a live agent.

Where the difficulty sits now, and what to test when you judge one.

Where a live agent earns its place in the B2B funnel.

A persistent spokesperson customers can talk back to.

Guardrails, disclosure, and what to test before launch.

What building a live agent actually costs in engineering time.

What makes an agent interactive rather than scripted.

The six stages between a spoken word and a rendered reply.

Five obligations to meet individually, not one checkbox.

What has to be true of a platform for that claim to hold.

An always-on concierge for the guest who arrives at 2am speaking any language.

Patients tell a face things they leave off a form.

A shop assistant who knows every product and is never busy.

The first face a new hire meets, available before day one.

A live demo and qualification on every pricing page visit.

Client onboarding and product education, inside the advice boundary.

The configurator at 9pm, when the showroom is shut.

Policy explained plainly, and a calm voice at first notice of loss.

Tier-1 support, billing disputes and the upgrade conversation.

Client intake after hours, without crossing into legal advice.

Tutoring and learner support at any hour, in the learner's own language.

Buyer qualification at every hour, without crossing into valuation advice.

Citizen information and service applications, multilingual and always open.

Trip planning with a face, and the disruption hour handled properly.

The five-minute window, and what happens when nobody answers it.

From audio to face movement, and where the hard part really is.

Data residency after Schrems II, and the question to put to a vendor.

Calibrating an allowance when every conversation costs real compute.

Practice conversations that score, instead of modules that complete.

Where the rooms are, who runs them, and how to get the most from one.