One person + a team of self-improving agents = 10×. Omniq Atlas gives every employee a team of specialist agents (20 role agents across developer, PM, director and CTO, with higher roles inheriting the agents below them). Those agents improve themselves every week by watching how people actually work. It solves the problem that enterprise agents ship once and never get better, and the risk that self-improving agents get worse or leak data. You hand off a job with @agent, and the agent answers in your context: your role, persona, projects, and only the documents IAM lets you see. There are no feedback forms. Rephrases, corrections, follow-ups (“now also write the rollback plan”, classified as an extension), regenerations and copies become signals in Atlas and in Raindrop. Raindrop’s Query API tells each agent where it fails, by failure type and playbook version. Every person and every agent runs a 7-day improvement cycle, and a Raindrop failure spike triggers one immediately. Each cycle proposes one change to the agent’s playbook (rules, retrieval, workflow or template). The change is kept only if it wins on unseen tests and a blind side-by-side replay, passes an absolute leak test, and reflects a pattern from 2+ people. Bigger changes wait for a manager. A watchdog compares Raindrop failure rates per version and auto-rolls back regressions. The gateway is model agnostic: local Gemma/Llama or any cloud model, with private data restricted to local models. Results so far: 26 automatic decisions, 6 kept, 12 undone, 4 refused, +38 points on kept changes, 0 leaks, and 78% of answers right the first time.
More from The Harness Engineering & Model Wrangling #SFTechWeek Hackathon
MongoMation is an autonomous PLC engineer. You describe a machine in plain English, and Claude agents write IEC 61131-3 PLC code, compile and deploy it to a…

LabRat is a self-improving harness for a physical lab agent, running on a real bench: two SO-101 arms, a 3018 CNC carrying a servo pipette I designed during…

Recursive Harness Scientist: an AI scientist that improves its own scientific method, and must replicate every gain on data it has never seen. Why it…

Super Bid is a self-improving AI estimator that teaches itself to price bids the way a specific company would. It practices on the company’s real past bids…

A robot that gets better at handling objects it’s never seen, without retraining. After every round, an AI coach reviews what broke and rewrites the robot’s…
WORLDSTATE gives physical spaces a memory. Any phone becomes an observer: open a link, point the camera, and WORLDSTATE tracks what's inside fridges, shelves…
