Twelve shipped things, most of them realtime, most of them on a free tier.
Server-authoritative game loops, CRDT sync over Cloudflare Workers, chunked storage on an attachment cap, on-device inference driving a canvas. The constraint is usually the interesting part.
You get the repository, the deploy pipeline and someone who can explain every decision in it.
That we will argue with the spec. You get people who ask what happens at ten thousand rows and who will tell you when the feature you asked for is a symptom of a different problem — which is the part a contractor is not incentivised to do.
TypeScript everywhere, React and Next.js on the front, Node or Python behind it, Postgres by default. We are unromantic about it: the stack should be the boring part, and the interesting part should be your problem.
Yes, and with a specific opinion about it. A model's output has to be *reviewable* before it is useful — that is the whole argument of Tessera, which puts an editing agent behind a document a person can read a diff of. Anything less is a feature you cannot debug.
In proportion. Types and a fast build catch most of it; tests go where the cost of being wrong is high — money, data loss, auth. We would rather ship a verification harness that checks the things that actually break than a coverage percentage.
Yes. Two or three of us inside your process, reviewing in your repository, is a common shape and usually the most useful one — you keep the context and we bring the hours.
You have everything, because you always did: your repository, your infrastructure, your accounts. There is no handover event because there was never a wall. If we did our job you should be able to keep going without calling us, and some clients do.