Lab 03 · Live board
Our AI agent, graded in public.
Arlo, our chat agent, answers visitors on every page of this site. This board is its report card: real production numbers, pulled live from the same worker that runs the chat. Not a screenshot, not a case study — and the accountable human it answers to is Puvin.
The numbers, as they stand.
Supervised by one accountable human.
No cherry-picking possible: this page reads a public endpoint, and you can too. Open it in a browser or a terminal — same numbers.
Every number, defined.
- Conversations
- Unique chat sessions that sent at least one message this calendar month (UTC). Opening the widget without typing doesn't count. Resets on the 1st.
- Resolved without Puvin
- The share of those conversations Arlo handled end to end. Two things count against it: any reply Arlo fails to deliver — logged automatically, because the system fails loudly to a human, never silently — and any conversation Puvin personally steps into, which he logs against the board when it happens.
- Median response time
- Server time from your message reaching our worker to Arlo's reply leaving it, recorded on every answered turn this month. Your own network adds a little on top.
- Last human escalation
- The most recent time the accountable human had to step in — because he chose to, or because Arlo dropped a reply. All-time, not just this month.
- Beta means beta
- Arlo is in Beta, and the numbers include its rough edges. If the resolution rate dips, you'll see that too. That's the deal with a live board.
- Humans involved
- One.
How supervision works day to day — reviewed, then spot-checks, then exceptions only — is in the FAQ.
Every site has a chat widget. Almost nobody shows you if it works.
The floating chat bubble is the most saturated pattern on the internet. Enterprise vendors publish their support agents' resolution stats at massive scale; studios ask you to take the demo's word for it. For a one-founder studio, the honest version is this: the meter, live, from production — including the months that look ordinary.
It's the same rule our client work runs on. The Shopify-to-AI-agent integration we built for Kenny Hills Hospitality Group replaced their manual monthly reporting entirely — in production and still running. The deal never changes: the agent does the repeatable work, and one accountable human answers for it.
Ask it something. It lands on this board.
Every conversation you start with Arlo becomes part of these numbers. Prefer the human it answers to? Fifteen minutes on WhatsApp finds the first task worth automating.