Confetti takes a plain-English description and gets hardware to your door. It does the CAD, finds off-the-shelf parts that actually work together, sorts the wiring and the firmware, then pulls quotes and orders the components for you. Two UPenn dropouts built it after trying to make themselves a home AI device and drowning in the stuff nobody tells you: which connector, will it fit in the case, who sells this, how long is shipping, does any of it match. You can remix other people's builds on there, and the code HUNT halves your first three months.

There's an app idea in everyone's ChatGPT history that never got built. Floot is now a native app in the ChatGPT sidebar, so it reads whatever you've already been talking about instead of making you re-explain it, shows a live preview in the chat while it works, and publishes to a real link or to both app stores. Database, logins and hosting come with it. It runs on the ChatGPT plan you already pay for, with no charge for tokens and 100 build actions a day. A camera rental firm in Czechia runs its bookings, payments and shipping on one.

Radar hands your coding agent a Kubernetes cluster that's already been worked out: the issues it found, what changed and when, how the objects relate to each other, logs, metrics. Without it an agent burns time and tokens running kubectl at the problem and still misses the deploy someone pushed ten minutes before everything fell over. Same agent on the same faults got to root cause three times faster and 80% cheaper. They put six AI SRE tools against 50 live faults and published every run, misses included, which is the only reason 44 out of 50 means anything.