Apr 11
Loading video player...
How to set up your codebase for running parallel coding agents like Claude code and codex cli. Full architecture guide with live demo. We dive into how to run ai agents in parallel, we run multiple claude code agents reliably and break down why it works. We compare codex vs claude code and much more. Most developers can't run AI agents in parallel because their project structure doesn't support it. Two agents touch the same file, merge conflicts everywhere. This video shows you how to fix that. In this video: → The 3-crate architecture that enables parallel work → Vertical slice pattern for AI-safe code structure → Unified naming: same grep pattern across code, logs, and GitHub labels → Live demo: 7 agents running in parallel (3 on same feature, 4 fixing bugs) → 3-way comparison: Codex vs Claude Code vs Kiro building the same UI Whether you're using Claude Code, Cursor, Aider, Codex, or Kiro - this architecture transfers to any tool. Shards supports: → Claude Code → Kiro CLI → Gemini CLI → Codex CLI → Cursor → Aider → Any CLI-based agent 🔗 Links: - Shards repo: https://github.com/Wirasm/shards - Work Tree Manager skill: https://github.com/Wirasm/worktree-manager-skill - Kiro Hackathon (free, $17K prizes): https://dynamous.ai/#/kiro-hackathon - AI Coding Course: https://community.dynamous.ai/checkout/dynamous-friend-discount?affiliate_code=d020a5 - Full playlist: https://www.youtube.com/watch?v=qFYCnzec_rg&list=PLQon6sL0PBjhniHswpI3_cRQstdx7Lmdo 📚 Kiro docs: https://kiro.dev/docs --- Timestamps below 👇 --- 🔔 Subscribe for AI coding architecture and parallel agent workflows. 0:00 - why most people can't run ai agents in parallel 0:30 - what we're covering today (architecture + live demo) 1:00 - the 3-way comparison: codex vs claude code vs kiro 1:30 - shards intro and series context 2:00 - kiro hackathon mention + honest review coming 2:40 - like and subscribe 3:00 - updating steering docs for the new crate structure 3:40 - the refactor from yesterday: 3 crates explained 4:00 - creating issue for per-provider trust mode flags 5:00 - workaround: updating the shards skill 6:30 - the PRD for the UI and phase 1-3 plan 7:00 - using cloud code to generate the plan (platform agnostic) 8:00 - the triage system and how labeling enables parallel work 9:00 - launching the 3-way comparison (kiro, claude code, codex) 10:00 - triggering bug fixes in parallel via arcon workflow 11:00 - kiro hangs again (ctrl+c continue workaround) 12:00 - why i'm done with kiro for this project 13:00 - architecture deep dive: vertical slices explained 14:00 - the 3 crates: core, cli, ui 15:00 - inside a slice: types, errors, operations, handlers, tests 16:00 - unified naming: code structure → logs → github labels 17:00 - same grep pattern everywhere (the unlock) 18:00 - how labels let you identify parallel-safe work 19:00 - optimizing for the agent's perspective 20:00 - prs created from all 3 agents 21:00 - running pr review agents in parallel 22:00 - the 3 parallelization patterns recap 23:00 - wrap up and next steps #aicoding #parallelagents #claudecode #aiagents #codingworkflow #agenticengineering