Real jobs people get done with AI agents, each tied to its original source, an honest outcome label, a prompt you can edit, and the GitHub repositories involved.
Anthropic split eight staff without robotics experience into two teams to program a quadruped robot to fetch a ball; the team using Claude finished more tasks and, on shared tasks, took about half the time.
DeerFlow 2.0 is an open-source agent harness from ByteDance that plans a task, hands parts to sub-agents and works in a sandbox with memory, skills and search tools, using any OpenAI-compatible model.
Google DeepMind's AlphaEvolve pairs Gemini models with automatic scoring to evolve code, and reports gains in data-centre scheduling, AI kernels and open maths problems.
Simon Willison sends research questions to cloud coding agents such as Claude Code for web and Codex Cloud, which write and run experiments in a dedicated repo and open a PR.
Red Hat's Roy Belio containerized Karpathy's autoresearch on OpenShift AI with H100 GPUs and let Claude Code run 198 training experiments over 24 hours.