🤓 Latest Submissions
Dark Factory: Build Blind, Review Hard
Most agent teams let the author check their own work. We banned it. Our dark factory is three seats in one Band room: a lead that coordinates, an implementer that writes every line, and a reviewer that is the only seat allowed to test. The implementer builds from the spec and never runs a single check, so every defect has to survive an independent reviewer who measures it, reproduces it and rejects it in the open. We pointed it at Pocketful, a wallet and payments service. After two dispatches and nothing else from a human, it delivered all four stages, each in its own buildable folder, passing the shipped checks in isolated mode. Six rejections changed the code. The best one: every shipped check was green when the reviewer's own tests showed that a new sign-up overwrote a seeded user and the wallets stopped adding up. No shipped check would have found it. Every rule in the factory comes from a recorded run: 18 of them, one shared monitor, append-only raw data. The whole four-stage run cost about 4.5 USD at list prices. What failed is documented too. The mandates are generic, so another team can aim the same factory at a different problem using FACTORY.md and the mandates folder.
6 Oct 2026
Shift Pilot — Voice-First Shift Handovers
In Colombia and across most of Latin America's industrial sector, shift handovers still run on paper notes and WhatsApp voice messages. Most plants have no digital system to capture what happened during a shift - an open alarm goes unlogged, a leak goes unflagged, and supervisors find out days too late. Shift Pilot replaces that gap. At the end of a shift, the operator just talks for 2 to 5 minutes. Speechmatics transcribes it, the operator corrects the text, and an AI agent (Featherless AI running DeepSeek-V4-Pro) extracts structured events and resolves every equipment mention against the plant's real asset catalog. When two assets could match - two near-identical pumps on the same line, for example - the agent doesn't guess. It stops and asks, using the catalog's real options as the only valid answer. The catalog also carries operator aliases, so a nickname resolves as reliably as a formal tag. Open events carry forward to the next shift automatically, by reference, never duplicated, with a shift trail tracking how long each one has been open. The incoming operator is blocked from starting their own shift until they've read the previous handover's summary and confirmed receipt. Supervisors get a dashboard with alerts ranked by how long events have stayed open, full shift history with no duplicated events, and PDF export. Built on React, TypeScript, and Supabase (Auth, Postgres with row-level security, Storage, Edge Functions), Shift Pilot was built end-to-end inside Natively - every screen, the database schema, and both edge functions calling Speechmatics and Featherless AI - without hairing integrations manually. Target users: plant operators who need to document a shift in minutes on a shared floor tablet, and supervisors who need a reliable view of what's open and who
10 Aug 2026
