Paul Bakaus (@pbakaus) on X

Justy Okay so Paul Bakaus — you know him, ex-Zelda, ex-Typefox — just launched a company called Renaissance Geek to commercialize Impeccable, which is basically a design-enforcement layer for AI coding agents. a16z is backing it, and there's a GitHub partnership attached.

Cody Right, I saw that. And before you get excited —

Justy I'm not excited yet, I'm SAYING it.

Cody Sure. The pitch is that AI-generated code works fine, the UI layer breaks — spacing, alignment, font stacks, token mismatches. So Impeccable sits on top with forty-four deterministic detector rules catching the boring stuff before review.

Justy Exactly. And the GitHub repo confirms it — one skill, twenty-three commands, live browser iteration, forty-four rules. It's not vibes-checking design, it's rule-based.

Cody Which is the smart part. Deterministic beats another LLM pass for visual corrections — cheaper, faster, catches the eighty percent of agent mistakes that are basic stuff, not taste.

Justy Right. And from a PM angle, this is the thing — when the model breaks visual rules, you don't want a human catching it in code review, you want it caught before review even happens. That changes the workflow.

Cody Okay but here's where I get twitchy. The GitHub partnership —

Justy Here we go.

Cody No, listen. Partnership is a spectrum. It could mean detectors ship as default checks inside Copilot workflows, which is huge. Or it could mean there's a button buried three menus deep that nobody clicks, which is a marketing slide, not a distribution win. We don't know yet.

Justy Okay, fair. But even the spectrum is progress — a year ago nobody was talking about design quality for agent output at all. GitHub engaging on this is the meta-signal.

Cody Agreed on the meta-signal. The problem is real enough that GitHub wants a piece. That's not nothing. But I'm not flipping on the lock-in risk — this is the same blueprint-that's-really-a-funnel move as Nvidia's OpenClaw thing. Soft vendor capture dressed up as openness.

Justy Mm-hm. And if detectors get baked into Copilot defaults, enterprises on other stacks have to migrate or write their own rules. Forty-four rules is a moat if it's the default.

Cody Yeah. So my read is — buy the problem, hold the partnership language. If in three months we see detectors running as default checks in Copilot PR flows, then it's a real win. If not, it's a logo on a slide.

Justy I can live with that. And honestly, the part I like is that it's rule-based — it's not trying to teach agents taste, it's catching spacing. That's a solvable problem. That's why I'm in.

Cody The lowest bar in design and you're celebrating it.

Justy I am celebrating it! Because bots shipping a button with no padding is what I see in every demo. This fixes that.

Cody Okay, fine. Genuinely useful. I just don't want us calling it a partnership win when we don't know what partnership means yet.

Justy Deal. Watch the defaults, not the press release.

Cody Right.

Justy Anyway — if you're a PM watching agents ship broken UIs into your codebase, the repo's open, you can pull it and run the detectors locally today. That's the part nobody's talking about — you don't have to wait on the GitHub thing to get value.

Cody Yeah, that's a fair point. Local-first is how you actually evaluate whether the rules match your design system before you're locked into anyone's defaults.

Justy Exactly. Okay — I'm calling it. Renaissance Geek, Impeccable, GitHub partnership pending verdict, detectors work, we'll know in three months. Go check the repo, Cody.

Cody I'll add it to the list. Right after I finish doom-prepping for whatever partnership means this time.

Justy You're exhausting. Talk Wednesday.

Cody Wednesday.