pi-cowboy is a Herdr-native orchestrator that breaks large coding jobs into focused subtasks and runs them as parallel background agents using Pi or alternative harnesses like Codex, Claude Code, and OpenCode. Each agent runs in its own Herdr terminal pane and works in an isolated copy-on-write Git worktree, committing changes to a dedicated branch so you can watch progress, jump into any agent to steer it, review reports when it settles, and merge individual branches into your main tree. The tool auto-detects available harnesses on PATH, supports reusable agent templates for recurring roles, and provides orchestrator templates to enforce cross-agent rules and reporting cues.
Operation is centered on spawn → isolate → settle → merge → cleanup. Agents queue when concurrency limits are reached (default global limit 4) and occupy slots at multiple levels (global, per-harness, per-provider, per-model). Cleanup is safe by default: it won’t remove running agents, won’t delete worktrees with uncommitted changes, and only deletes branches after merge; a forced removal exists in the UI with explicit confirmation. Requirements include Linux or macOS, herdr >=0.8 with Pi integration, python3 for macOS, and a copy-on-write filesystem (APFS, BTRFS, ZFS) to keep disk overhead small - e.g., a 153k-file repo showed 75 MiB for pi-cowboy’s COW worktrees versus 479 MiB for classic worktrees and 27 GiB when including untracked files. Installation is via npm (pi install npm:pi-cowboy).
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.
Nicholas Charriere uses a camera and AI to create a daily habit of practicing math and physics, with an AI coach providing exercises and grading. This approach helps him regain his skills and enjoy using his brain again. (twitter.com)
Superlite is an autonomous build workflow for Claude Code that splits tasks to run simultaneously, significantly reducing wait times. It automates planning, execution, and review, allowing users to focus on higher-level work instead of constant supervision. (contextfirst.blog)
AI agents are being developed to trade stocks automatically while investors are asleep. Charles Schwab's CEO expresses skepticism about the effectiveness and reliability of these AI-driven trading systems. (wsj.com)
A prediction market asks whether a major-party U.S. presidential candidate will deploy an AI clone to communicate with voters before the 2028 election. The AI system must represent the candidate and generate real-time, substantive responses, with deployment authorized and verified by the campaign. (manifold.markets)
![Richard Feynman: Can Machines Think? (1985) [video]](https://hntoday.pixelshot.io/screenshots/20261019/9276abea066c4acb_tn.jpeg)
Richard Feynman explores the question of whether machines can think, discussing the nature of intelligence and computation. The video delves into the philosophical and scientific implications of artificial intelligence. (youtube.com)
OpenFlow is a visual builder designed for creating and managing multi-agent AI workflows using the opencode engine. It aims to streamline agent orchestration, workflow automation, and AI tool integration. (github.com)
Today's best Hacker News stories, summarized and screenshotted, one email a day.