I Built the Ultimate Multi-Agent Workflow w/ Hermes Agent Kanban Board

I built a fleet of Hermes agents that scout real problems, research them, and build the fix — fully autonomous except for one human approval. Full workflow open-sourced at https://github.com/tonbistudio/hermes... Sign up for my FREE weekly newsletter, where I spill my unfiltered thoughts on the latest AI news, cool research, and projects I'm building: https://www.onchainaigarage.com/ 🐦 Follow Tonbi on X for real-time AI x blockchain updates! https://x.com/tonbistudio Everyone talks about multi-agent workflows; almost no one shows one running end-to-end without breaking. This is the real thing, built on Hermes Agent's Kanban board. I break down why coordination — not the agents themselves — is the hard part, and how the Kanban solves it: every unit of work is a card in one SQLite file (the single source of truth, no message queues), a dispatcher loop claims ready cards and spawns the assigned agent in a clean workspace, and tasks auto-promote themselves through parent/child dependencies so work fans out to parallel researchers and fans back in with zero glue code. The result is durable (survives crashes), parallel, event-driven, self-healing, and fully auditable. Then I run my own workflow live: paired scout agents (an X scout on a Grok model + a web scout on Reddit/YouTube) hunt real pain points AI-agent users hit, an orchestrator dedupes and scores them against a rubric (anything under 65/100 gets shelved), three researcher agents verify each survivor in parallel, and the orchestrator decides whether to build a tool, make a video, or shelve it — routing the winners to me in Telegram for the one and only human gate (approve / shelve / modify). Approve a build and a builder + tester ship a working CLI; approve a video and a producer hands back slides + a script. In one run you'll watch 18 agents work in parallel, hit 97 completed tasks, see the system self-heal a broken deliverable on its own, and walk away with a real CLI tool and a video deck. Then I open-source a generalized version of the whole template so you can adapt it to your own work. Resources: 🔗 Hermes Multi-Agent Workflow template: https://github.com/tonbistudio/hermes... Timestamps: 0:00 - What I built: an autonomous agent fleet 1:16 - The Hermes Kanban board and why coordination is the hard part 3:57 - How a card becomes a running agent: dispatcher + auto-dependencies 6:26 - Why this wins: durable, parallel, event-driven, self-healing 7:05 - The workflow: pain-point scouts to build-or-video, one human gate 12:07 - The scoring rubric and the shipping bar 12:55 - Live run: scouts, orchestrator, 18 parallel researchers 21:48 - Approving a build and a video proposal in Telegram 25:48 - Self-healing recovery in action 28:30 - The deliverables + open-sourcing the template What kind of multi-agent setup are you running — or trying to? What workflow would you point a fleet like this at? Drop it in the comments. I'll be doing a lot more in the agentic-workflow space, so if this was useful please like, subscribe, and hit the bell! 🦐✨ #HermesAgent #MultiAgent #AIAgents #Kanban #AgenticWorkflows #Automation #LocalLLM #AITools #VibeCoding