This AI Agent Builds Its OWN Tools to Finish Tasks! (INSANE Agent S4 Test)

Agent S4 is a new research system where an AI agent builds its own tools instead of clicking cell by cell. I explain the paper, then race click-by-click against a generated tool on my own laptop. This is my small demo of the idea (8B local model, RTX 4060 8GB), NOT a reproduction of the paper's benchmark. Benchmark numbers are author-reported. Chapters 0:00 The problem: cell by cell 0:29 What Agent S4 is (repo, release status) 1:01 How the interfaces evolve 1:32 The reported numbers 2:09 My demo setup 2:28 Click by click (20 cells live) 2:58 The tool the model wrote 3:18 An unseen spreadsheet (and a crash) 4:06 What this shows and what it cannot Sources Paper: https://huggingface.co/papers/2610.28601 Code: https://github.com/simular-ai/agent-s (agent_s4 folder, on main; latest tagged release is v0.3.2 from Dec 2025) If you are looking for cheap and reliable GPUs, I use RunPod: https://get.runpod.io/pe48 Disclosure: the avatar intro/outro clips are AI-generated from my likeness and cloned voice. Get GPUs Runpod: https://get.runpod.io/pe48 Get Hostinger: https://www.hostg.xyz/SHJiu Coupon Code "PROMPT" 10% discount 🔗 Connect with me: Email → [email protected] Ko-fi → https://ko-fi.com/promptengineer Patreon → https://www.patreon.com/PromptEngineer975 Book a Call → https://calendly.com/prompt-engineer48/call GitHub → https://github.com/PromptEngineer48 Twitter/X → https://twitter.com/prompt48 Website → https://promptengineer48.com/