Stop Chaining Your AI Workflows. Build a Board Instead.
A multi-agent kanban board just shipped. Here's the operator-grade pattern you can steal even if you never touch the tool.
Just spent the last week building a 30-day post-purchase sequence in GoHighLevel for a new product launch.
24 touchpoints over 30 days. Welcome email, install help, skill-by-skill education, soft pitch into the next tier, all of it. Standard work. Took a couple of long sessions to get right.
Then I had to build three more workflows.
Not because the sequence wasn’t working. Because the platform has no clean way to PAUSE a contact in the middle of a sequence and resume them later. So when a buyer upgrades to the next tier, books a clarity call, or hits a refund window, I needed a side workflow to catch the tag, remove them from the main flow, and in one case re-add them seven days later.
Four workflows to do what should be one.
That is the daily reality of building automations on operator-grade tools right now. The chain is the only primitive. You go from step one, to step two, to step three. If anything happens off the rails, you handle it as a separate workflow firing on a separate trigger.
Then yesterday, Nous Research shipped something that breaks that pattern wide open.
It is called Hermes Agent v0.12.0, and the headline feature is a Kanban board for AI agents. Tasks go on the board. Agents claim tasks. Agents work on tasks in parallel. Agents hand off to other agents when blocked. You can drop a comment on any task at any time and the next worker picks it up. Everything is logged to a database file on disk, so a crash does not lose state.
The operator version of that sentence is short: imagine if your nurture sequence was a board, not a chain.
What just shipped
The release is called Hermes Agent v0.12.0, “the Curator Release.” Here is what is in it:
A SQLite-backed task board you can run on your laptop or your server.
A live dashboard showing every task and its status (todo, ready, running, blocked, done).
Multiple AI agents running as separate processes, claiming tasks off the board, working in parallel.
A comment thread per task that humans and agents both write into.
Heartbeat monitoring so workers can signal “still alive” during long jobs.
Crash recovery: if a worker dies, the dispatcher reclaims the task and another worker picks it up.
Slash commands from inside any chat surface (Telegram, Discord, Slack, your CLI) to read, write, or unblock tasks.
That last one is the part operators should sit up for. You can be on your phone, away from your laptop, see a notification that a task is blocked because the AI hit something it does not have context for, and unblock it with a one-line text message. The next time a worker spawns on that task, it reads your comment as part of its context.
The mindset shift
A practitioner blogging about Hermes this week put it into one sentence:
“Multi-Agent Kanban changes your role from worker to manager. With a board, you can start assigning outcomes instead of controlling every step.”
Three things change when you flip from chain to board:
You stop deciding the order. In a chain, you write the if-then-else in advance. In a board, the dispatcher promotes any task whose dependencies are done, and whichever agent has capacity claims it. The order emerges from the work, not from your spec.
You stop losing state. Every handoff is a row in a file on your disk. Crash the worker, kill the laptop, restart the dispatcher. The work resumes from where it stopped. You do not lose progress because a process died.
You stop being the bottleneck. A chain that needs human input stops cold until you are at your desk. A board lets a worker post “blocked: need decision on rate-limit key” and keep all the OTHER workers running. You unblock from your phone when you get to it.
Why this matters for operators
I do not expect a marketing consultant or an agency owner to install Hermes Agent and wire up SQLite tomorrow.
That is not the lift. The lift is recognizing that the pattern, the board with claims and handoffs and human-in-the-loop, is the same pattern you can build on top of n8n, Make, Zapier, GoHighLevel, or any operator tool you already pay for. The pieces all exist. They have just not been wired up the way Hermes wired them up.
Three of the eight collaboration patterns in the Hermes docs are the ones operators should steal first:
Fan-out, fan-in. You drop one big request, the system kicks off three parallel research tasks, all three feed into one synthesis task. Today most operator tools do this in series. The board version does it in parallel and finishes in roughly one third of the time.
Pipeline with gates. Scout finds the leads, editor cleans the data, writer drafts the outreach, reviewer approves before send. Today most automation tools wire this as a single linear flow with approval steps tacked on. The board version lets the reviewer hold a card, comment on it, and pass it back to the writer for one rev without rebuilding the whole flow.
Human-in-the-loop. This is the one I want most for operator workflows. Every chain I have built in the last year has at least one place where the right answer is “ask the operator and wait.” Today that breaks the chain. The board version lets a worker post a question, freeze just that task, and keep the rest running.
The compression thesis
If you have been paying attention, the agency stack of 2024 is getting compressed.
The dedicated retention tool is becoming a feature inside your CRM. The dedicated lead-research tool is becoming a feature inside your AI assistant. The dedicated workflow orchestrator (Zapier, Make) is becoming a feature inside your model context. And now, the dedicated multi-agent project management layer is something you can run on your laptop with a SQLite file.
The pattern is not subtle. Every layer of the agency stack is being absorbed into the AI layer of the tools operators already pay for. The board pattern is the same story applied to multi-step automation.
Operators who recognize this in May 2026 are going to spend the rest of the year compressing four-workflow setups into one and freeing up the cognitive load they have been spending on “what order should this run in” toward “what outcome am I trying to produce.”
What to do with this in the next seven days
You are not going to install Hermes Agent. You probably do not need to. Here is what I would do this week if I were running operator-grade automations and wanted the board pattern without the SQLite.
Pick one chain that already exists in your stack. Does not matter which tool. n8n, Make, Zapier, GHL, all have a flagship workflow you have been duct-taping for a year. Pick that one.
Map every stop point in that chain. Every place a human currently has to step in. Every place that breaks if a step takes longer than 60 seconds. Every place where you wish a different specialist could pick up and the rest could keep moving. Those are the places a board pattern would help.
Replace the linear chain with a queue table. This is the operator-grade move. You do not need SQLite. You need a Google Sheet, an Airtable, a Notion database, anything with rows and a status column. Status is column one. Assigned-to is column two. Outcome is column three. Now your “chain” becomes “look at the next row whose status is ready and run it.” You have just rebuilt the kanban primitive on top of your stack.
You will discover within a week that 60% of your automations were doing things the chain only does because the chain is the only thing your tool gives you. Once the chain is replaced with a queue, you can fan out, you can route to a different agent based on the row, you can pause one row without breaking the rest, and you can hand off to a human without rebuilding the whole flow.
That is what just shipped. Not a tool. A pattern. The tool is one implementation of it.
If you want a hand mapping your flagship chain into a board pattern, book an AI Clarity Call at muddventures.com/book. 30 minutes, no pitch, free, you walk out with a one-page audit of where the board pattern would save you the most time in your business.
And if you want more breakdowns like this one, what just shipped, what it means for operators, and the templates to actually deploy it, the Abra AI community at whop.com/abra-ai is where I drop these patterns first, alongside the skills, templates, and the group of operators running them.
Andrew Mudd
Mudd Ventures