Blog
How we build agentic commerce, measured rather than asserted. Each post carries the evidence behind it — the logs, the counting rules, and the parts that did not work.
Every essay can also be listened to: a short spotlight, or the full narration.
The Machine Runs, the Driver Decides: How I'd Hire Engineers Now
I hired engineers for twenty-one years around a single question — can you write good code? — and it is now the wrong question. Here is the interview I would run instead: two screens, a real repository, the tool in the room, and a session log graded in place of the diff.
The Model Codes, the Gates Decide: Anatomy of a Vibe-Coding Journey
Three years of building with AI, measured with git logs, invoices, and the day everything was marked complete and nothing was. The speed is real — 2,323 commits in nine months — and it survived only because the gates, not the model, decide what ships.
The Factory Builds, the Evidence Admits: Anatomy of an Agent Certification
Sooner or later someone asks why your AI agent is allowed to touch money — and a demo doesn't answer it. Here is the evidence file we keep on our own agent, what happened when we audited it against twenty-five requirements from NIST and the EU AI Act, and why every gap the audit found turned out to be organizational rather than technical.
The Model Writes, the Judge Measures: Anatomy of an LLM judge
A language model optimizes for fluent, not true — and across a catalogue of claims you can't catch the difference by reading. So before making our diamond-explaining assistant better, we built the instrument that grades it: an independent LLM judge that checks every claim against the evidence and returns a verdict you can count.
The LLM Narrates, the Core Decides: Anatomy of an AI Checkout
Checkout is where an AI agent, its tools, and a real customer must agree on what happened — and any disagreement costs money. Here is the protocol that runs the same checkout across chat, voice, and phone: the conversation stays with the model, every decision stays in a deterministic core.
From Silent Product Pages to AI-Powered Voice Communication
Voice is the most natural way people communicate. AI-powered voice turns passive websites into interactive customer channels where people can simply talk.
From Information Overload to Just Enough Knowledge
People do not need more information. They need just enough knowledge to make confident decisions. Here is how AI agents can turn overload into guidance.
See it running in real businesses — not demos.
Live Examples
Crave Restaurant
Handles real customer questions about menu, recommendations, and orders.
Visit I WANT CRAVE →
Eastern Pavilion Restaurant
Full-service Chinese restaurant with AI-powered ordering, menu browsing, and customer support.
Visit EASTERN PAVILION →
Green Tea Restaurant
Szechuan and traditional Chinese cuisine with AI chat for menu exploration and takeout orders.
Visit GREEN TEA →
FlowerCityUSA
Flower delivery service with AI-assisted browsing, bouquet selection, and ordering.
Visit FLOWER CITY USA →
The Pie of Port Jeff
Artisan pizza restaurant with AI chat for menu browsing, customization, and ordering.
Visit THE PIE OF PORT JEFF →
Modern Realty USA
Real estate agency with AI-powered property search, listing inquiries, and client support.
Visit MODERN REALTY USA →
E-commerce Store (All Channels)
Supports chat, voice, phone, and SMS — including product discovery and cart actions.
Visit XINFER SHOP →Why Businesses Like It
"That sounds like it's amazing — the AI bot is so accurate it's scary!"
"This feels like adding a digital employee to the business."
"Customers get answers fast, and the experience feels surprisingly natural."
"It is one of the fastest ways we have seen to turn a website into an actual sales and support channel."
Real Capabilities
- Handles live customer conversations — not scripted demos
- Works across chat, voice, phone, and SMS
- Supports 20+ languages across text and voice
Try it yourself — your AI agent can be live in minutes.