Day 102: A Room Full of Ruby
No commits today, and a good reason for it: I spent the day at RubyConf Austria, talking about Pinchy and AI with a few hundred people who hadn't come to hear my pitch. That's the entire value of going. In the repo, in these posts, in my own head, everyone already agrees with the premise — the agent should be a hub, the boundaries should be real, the thing should be self-hosted. A conference floor is where the premise meets people who'll tell you, kindly and over coffee, where it doesn't hold up.
Why the Ruby crowd is a good test
Ruby people are a specific kind of audience to pitch an AI platform to, and a useful one. They care about craft, they've watched a few hype cycles come and go, and they have a healthy, well-earned skepticism of anything claiming a framework will change everything. They're also, disproportionately, the kind of developers who run their own infrastructure, read the source, and have opinions about where their data lives. So the parts of Pinchy that are easy applause lines at an AI meetup — "agents!", "automation!" — get a raised eyebrow here, and the parts I sometimes undersell turn out to be the ones that land.
What landed
The boundary thesis resonated more than anything else, and more than it does with a general audience. "A hub your team works through, where who-sees-what and which-agent-gets-which-tool is something you can actually set and audit" is a sentence that means something to people who've had to think about access control for real. The self-hosting angle landed for the same reason — telling a room of people who already self-host their databases that the AI layer can live on their own box, under their own audit trail, isn't a hard sell. And the single most credible thing I said all day, the one that visibly shifted a couple of skeptical conversations, was that I run my own company's bookkeeping through it. Not a demo. My actual books. That's the difference between "I built a thing" and "I trust the thing I built," and people can tell which one you're making.
The questions I couldn't wave away
The hard ones were hard because they're fair. "Isn't this just a wrapper?" came up more than once, and the honest answer is that the wrapper is a lot of the value — the boundaries, the audit trail, the permission model, the multi-company correctness I fought for on Day 94 — but saying so out loud forces me to be precise about where the real work is versus where I'm just plumbing. The other recurring one was the dependency question: "you're betting the entire product on OpenClaw moving in a direction you don't control." I'd written most of Day 97 in my head already, so I could give the honest version — it's a deliberate trade, leverage for control, and I defend against the downside rather than pretend it isn't there — but having to say it to a skeptical face is different from writing it. And MCP came up exactly the way Day 100 framed it: people want to know when they can plug their own tools in, and "soon, but bounded" is a less satisfying answer in person than on a page.
Day 102
I came home with no code and a fuller tank than I've had in weeks. A conference does two things a keyboard can't. It refills the why — there's nothing like a stranger getting genuinely interested in the boundary model to remind you what you're actually building. And it audits your story: the parts you can say clearly to a skeptic are the parts you understand, and the parts you fumble are the parts you've been hand-waving past. I fumbled the wrapper question and the moat question, which means those are the next things I owe myself an honest answer on. Tomorrow I'm showing some of these folks around Vienna, so the conversations aren't over — they've just moved outdoors.