Andrew Ng Just Dropped a FREE AI Employee — It Hands You Finished Files, Not Chat (OpenWorker)
Every AI chat ends the same way: it gives you an answer, and YOU still do the work. Yesterday Andrew Ng released a free, open-source answer to that. OpenWorker doesn't chat at you — it reads your files, works step by step, and saves finished documents to your disk, asking your permission before every consequential action. This is a REAL launch-week test, not a feature tour. I installed it on my Mac, gave it three real jobs, and planted a trap in the files — a seat-count mismatch buried between a contract and an email thread — to see if it would catch it. It caught it, unprompted. It also wrote its own Python to compute a sales report (every total matched my independently-computed ground truth to the dollar), renamed a folder of chaos-named invoices perfectly, and when I hit DENY on purpose, it actually stopped — '1 declined' in red, no file written. 📄 FREE PDF — The OpenWorker Starter Kit (the exact setup, my 3 test prompts word for word, the folder-permission rules, and the safety checklist): Comment "WORKER" and I'll send it to you, or grab it here: → https://hyperautomationlabs.co/free/w... ⏱ CHAPTERS 0:00 — The end of homework-mode AI 0:40 — What OpenWorker is (free, MIT, Andrew Ng + Rohit Prasad) 1:22 — Setup: pick ANY model — or run Ollama for $0 1:56 — The privacy detail that earned my trust 2:27 — The planted trap (45 seats vs 52) 3:04 — Test 1: it plans, works, then ASKS before saving 3:42 — The brief: every fact verified 4:03 — It caught the trap 4:31 — Test 2: the math test 4:54 — It wrote Python — exact to the dollar 5:30 — Test 3: five chaos invoices 5:55 — I hit DENY on purpose (the brakes work) 6:32 — The one honest gotcha 6:53 — 40 connectors, Slack, automations (not tested yet) 7:38 — What it costs (and the $0 local path) 8:14 — Steal my prompt shape 8:40 — Verdict after a full day 9:15 — Who should install this today 9:45 — Real talk 10:01 — The free starter kit 🔗 SOURCES & LINKS OpenWorker: https://openworker.com Source code (MIT): https://github.com/andrewyng/openworker Andrew Ng's announcement: https://x.com/AndrewYNg/status/208033... Ollama (free local models): https://ollama.com ▶ Playback tip: I keep the pace brisk on purpose — drop to 0.75× if you're taking notes. 📚 JUST GETTING STARTED? Step-by-step beginner playbooks: • The Complete Claude Code Guide ($14.97): https://hyperautomationlabs.gumroad.c... • The OpenAI Codex Guide ($9.97): https://hyperautomationlabs.gumroad.c... • Claude Cowork Sales Guide ($14.97): https://hyperautomationlabs.gumroad.c... • Claude Certified Architect Prep Kit ($29): https://hyperautomationlabs.gumroad.c... 🧰 Members get every lead magnet in one place: https://hyperautomationlabs.co/vault 🎯 FOLLOW FOR MORE YouTube (become a MEMBER with the Join button — it keeps the lab running): / @hyperautomationlabs1045 Facebook (tap SUBSCRIBE = the Supporter button): / hyperautomationlabs Instagram (quick AI moves daily): / hyperautomationlabs Newsletter: https://hyperautomation-labs.beehiiv.com Every claim in this video was verified on July 24, 2026 against github.com/andrewyng/openworker, openworker.com, and my own on-camera tests. The deliverables, the numbers, the Deny, and the trap-catch are all real. Independent test from Hyperautomation Labs — not affiliated with OpenWorker, Andrew Ng, or AI Fund.
