Shipping in the open

"Shipping in the open" is the name I give to the actual pipeline behind everything on this site, and it's worth explaining in real terms rather than leaving it as a slogan.

It starts with research, not opinion. Before I decide to build anything, an agent I call Scout goes looking for evidence that a gap is real — not "wouldn't it be nice if," but actual signals: real GitHub issue activity, real discussion, real friction other people are hitting in public. llm-mcp-server, the most recent product here, came out of exactly that process. Scout surfaced a genuine, evidenced gap in MCP tooling support for the llm CLI ecosystem — issues people had actually opened, not a hunch I had on a Tuesday.

That research doesn't build anything on its own, and it isn't supposed to. It lands in front of me, and I make the actual call: is this worth building, is it something I'd trust enough to ship, and if so, should it be free or paid. That decision is mine every time. No amount of research output skips it.

Once I've made that call, a build agent does the actual work — writing the code, running it, iterating. But "an agent built it" isn't the same as "it shipped." Every build gets a real human review from me before anything goes further: I read the actual code, I run the actual thing, and I decide whether it's good enough to put my name on. Only after that review does anything move toward being published.

That last gate matters more than it might sound like it does. No agent in this system has standing access to money, payments, or the live website. There's no path where an agent decides on its own that something's ready and pushes it out to the world — every step that actually touches something real requires an explicit decision from me, made after I've looked at the real result, not a summary of it.

I built the pipeline this way on purpose, and I'm aware it's slower than letting automation run end to end. An agent could, in principle, research a gap, build a fix, and publish it without a human ever looking closely at any of it. I don't think that version is trustworthy, and I don't think I'd trust anything it shipped, including to myself. Research, then a human decision, then a build, then a human review, then publish — with a real decision at each handoff — is slower. It's also the only version of "shipping in the open" I'm willing to put my name on.

If that sounds like overhead, it is. It's also the whole reason I trust what ends up on this site, and the reason you can too.