Five beliefs about software, AI,
and what we’re willing to fight for.
Most AI product manifestos read like marketing. This one tells you what we’ll argue with you about and what we won’t.
Installable software is not dead. Cloud-only is not the endgame.
For fifteen years, software product strategy has assumed every app should be a SaaS. SaaS reduces piracy, enables continuous deployment, lets vendors lock in customers. So far, so consensus.
But step into an accounting office, a clinic, a plumbing dispatch room. The software running their actual business is desktop. The receipts printer requires a Windows driver. The audit log lives in a folder, not a cloud bucket. SaaS hasn't reached them — not because they're backward, but because their workflows depend on physical hardware and locality.
The next generation of AI app builders are all web-only. Lovable, v0, Bolt, Replit Agent: every one of them stops at "deployable web app." That ceiling shuts out half the economy.
We don't accept that ceiling — but we haven't crossed it yet either. Today the factory ships verified web apps. The Windows desktop bridge architecture is in the repo; no desktop EXE has been built end-to-end on prod. Same for mobile. Roadmap, not shipped.
Proof beats demos. Every claim needs an evidence URL.
The AI app builder space is drowning in launch videos. Slick, edited, cherry-picked. Everyone's "AI agent" produces beautiful Twitter clips and never the actual installer.
Our position: if you can't hand a buyer a downloadable artifact with a SHA-256 hash, you don't have a product. You have a demo reel.
Every passing factory job produces a downloadable .tar.gz with a SHA-256 hash. The live list lives at /factory/proof — server-rendered from sf_factory_jobs on every page load. Passes AND failures. No curation.
When competitors stop showing demo reels and start linking artifacts, we'll know they're serious.
A factory is not an assistant. It produces, you don't supervise.
Cursor, Copilot, Windsurf — these are assistants. They make a developer faster. They require a developer.
A factory is different. A factory takes a description of what you want and produces it, autonomously, with proof. The operator doesn't open the code. The operator opens the artifact.
This distinction matters because the market for "developer assistants" is a few million developers worldwide. The market for "I'm a bookkeeper / clinic owner / dispatcher who needs this software and can't hire engineers" is a few hundred million businesses.
We build for the second market. Devin builds for the first. Lovable straddles. We're fine with that.
Multi-model is the answer. Single-vendor lock-in is the question.
In 2026, model quality is a moving target. Claude wins this week on long-context reasoning. GPT wins on coding. Gemini wins on multimodal. Mistral wins on cost.
Locking your factory to one provider is choosing whose bad weeks you live through. You take their outages, their rate limits, their pricing changes, their training-data drift.
Our gateway runs twelve models across four providers, switchable per request. Cost-sensitive runs go to Mistral. High-accuracy runs go to Claude Opus. The user picks. We don't hide the choice behind a "magic" router that secretly costs more.
The honest truth: every wrapper-of-one-model is a wrapper of one CEO's mood.
Honesty compounds. Hype erodes.
We've seen AI products launch with bombast and quietly walk back claims six months later. SOC 2 "soon." F500 logos on a wall those companies never authorized. Demo reels that won't reproduce.
Our calibration: if something is shipping, we ship it. If something is in progress, we say so plainly. If something is just "considering," we say that too — and ask whether you'd pay for it.
We have a working web pipeline producing downloadable apps (see /factory/proof). Zero signed design partners. Zero F500 logos. SOC 2 Type II in progress, not certified. Desktop and mobile pipelines are roadmap. We'll list real customers when we earn them. We'll claim certifications when we receive them. We'll match competitor features when we ship them — not when we say we will.
Hype gets you the first deal. Honesty gets you the renewal.
Do you want to build this with us?
If anything here resonated — especially the part about ditching demo reels for proof — we want to hear from you.
Or argue with us at [email protected]. We’ll listen.