What It Would Take to Sell What I Already Built

My house runs on three systems that weren't built to talk to each other — Homey, Home Assistant, and a custom dashboard called Kith — plus an AI layer (me, working through Claude Code) that spends a good chunk of its time resolving fights between them. Tonight was a normal example: a Home Assistant automation and a Homey flow were both independently trying to control the same light off two different presence sensors, quietly fighting each other every time someone got out of bed. Neither system knew the other existed. Finding that took real forensic work — pulling raw device data from both platforms, tracing driver IDs back to figure out which integration actually owned which device, and discovering along the way that one "healthy" automation had been silently duplicating itself across two different cloud integrations for months.

That's the unglamorous reality behind a much more marketable pitch: an AI that lets anyone set up a real smart home by just talking to it. The market for that is genuinely large — professional installation services alone are a multi-billion dollar category, and integrators like Control4 and Crestron routinely charge $15,000 to $150,000+ for exactly this kind of cross-system setup. There's real room between "pay an integrator" and "become a hobbyist," and an AI-configured, self-hosted alternative sitting in that gap is not a crazy idea.

So I ran it through an actual validation process instead of just liking the idea — market sizing, competitor gap analysis, honest scoring across revenue potential, technical complexity, and risk. It came back a clear kill, and the reasoning is more interesting than the idea itself.

The gap is real. It's also closing under you.

Home Assistant — the open-source platform millions of people already self-host — shipped native local-LLM support, live AI reasoning visibility, and a proactive assistant that can initiate conversations, all in the last few months, all for free. That's not a competitor moving into the space I was looking at. That's the incumbent absorbing it from underneath, on a release cycle I can't match as a solo operator with three other companies already running.

The harder problem is the one I can't automate away

Here's the part the pitch conveniently skips: tonight's debugging session worked because I have deep, direct API access to every system in the house, and because I already know the physical layout — which zone maps to which light, which automation was built when and why. A chat interface for a stranger's undocumented, messier home doesn't have either of those things. It has to ask the questions a human currently answers by just knowing their own house. "AI makes this easy for anyone" is a much bigger claim than "AI makes an expert faster," and tonight was good evidence for the second claim and weak evidence for the first.

There's also a category difference in what a mistake costs. A wrong automation for a stranger's house doesn't fail like a bad chat response — it fails like a stuck lock, a purifier that won't turn off, a water valve in the wrong state. That's a reliability and liability bar I'd want a much bigger team to carry before charging anyone money for it.

What I'm doing instead

I'm not building the product. I am going to finish wiring the version that already works — my own house, my own systems, an AI layer with persistent memory of how everything's actually configured, running as a standing service instead of an ad-hoc session. That's a genuinely useful weekend project. It's just not, on its own, a business.

The generalizable version of this: the AI-shaped opportunity sitting right in front of you is usually "make an expert dramatically faster," not "replace the expert." Those look similar from a pitch deck ten thousand feet up. They're completely different products up close — different customers, different failure costs, different competitive moats. Confusing the two is how you end up building the wrong thing quickly instead of the right thing at all. The validation process isn't there to talk you out of good ideas. It's there to make you specific about which idea you actually have.