AI Profit Boardroom· the Agent OS playbook Join the Boardroom →
Hermes Agent · Multi-Agent Kanban

Your Hermes Kanban used to freeze.
Now it runs a whole team.

Run a whole board of AI agents at once — a Planner, Builders, a Reviewer, all working together — and it doesn't lock up anymore. One slow task used to freeze the entire Kanban. The new Hermes update fixes exactly that. I ran a real content team on my Kanban to prove it.

See the real run ↓ A glowing pulse that keeps beating while heavy work is handed off to background helpers
The big announcement

Hermes just shipped this

This dropped from Teknium — cofounder and lead engineer on Hermes. It's the whole reason for this guide:

Let me translate it, line by line

"scaling huge multi-agent kanban runs" — this is the headline. The one thing that used to break — running a whole board of agents at once — now just works. That's the exact content factory I'll show you below.

"large concurrent sessions" — you can have loads of agents (or chats) going at the same time without it choking. More agents no longer means more freezing.

"this new update to database concurrency" — that's the fix in one line: slow saves stopped freezing everyone. I explain exactly how, in plain English, right below.

"Install or update now and give it a try" — it's already live. Not a roadmap promise — you can run it today. The demos and my live run below are me doing exactly that.

View the original post on X →

You don't need a smarter AI. You need agents that don't trip over each other.that's all this update fixes — and it's huge.

In plain English first

What this actually means — no tech words

Before any of the geeky stuff, here's the whole update as a simple picture.

Imagine you hire 10 AI workers to run your business. They all share one assistant, and that assistant writes everything down in a single notebook.

Here's the problem you kept hitting. Every time one worker needs something saved, the other nine have to wait their turn at the notebook.

And if one save is slow, everyone freezes until it's finished. So the second you tried to run more than one or two agents, the whole thing jammed up.

This update changes one thing. The assistant now hands slow saves to a back-office helper — so your workers never stand around waiting on each other.

The result: you can finally run a whole team of AI agents at once — a content factory, a research team, a whole agency of agents — and it doesn't freeze. That's the update. Everything below just shows it.

My story · why this matters

I was you. Then the freeze stopped happening.

Before

I tried to run more than a couple of agents at once and the whole thing crawled.

One agent would hit the database to claim a task, and every other agent just… waited.

A big Kanban run would stall halfway and I'd have no idea which agent was stuck.

Two agents would grab the same task and do the same work twice — burning tokens for nothing.

The breaking point was watching a 20-agent job freeze solid on a single slow write.

Then Hermes shipped the database-off-the-event-loop update.

After

I gave my Agent Kanban one topic this morning.

A team of Hermes agents planned five articles, wrote every one, and shipped them live to my site.

Nothing froze, nothing doubled up, and the board stayed responsive the whole time.

That's the difference between a demo and a thing your business can actually lean on.

You can run this exact setup today. Same tools. Same path.

III · the receipts

Real operators. Real fleets. Same result.

I'm not theorising about agent infrastructure. This is the stack 3,600+ founders inside the Boardroom are running right now — agency owners, ecom founders, course creators — across 38 countries.

400kYouTube subscribers
3,600+Boardroom founders
163kX / Twitter
38countries live
29kUdemy students
Before you scroll on —

Commit to transitioning today. Not someday.

You've seen what changed. The next few minutes show exactly how it works and exactly what I built with it.

So here's the deal I'm making with you.

Promise yourself one thing right now: you'll finish this guide and run one agent job before you sleep tonight. Just one.

Because the moment your agents stop freezing, the whole shape of what you can build changes.

The people sitting still are getting passed by the ones who are shipping fleets of agents today.

Be one of the people who moved. Commit to the transition. Take one action today.

The framework

The Unfreezable Loop

Every agent system that scales follows the same five moves. This update nails all five. Steal it for your own builds.

1

Find what freezes your agents

One slow save makes every agent stop and wait. The first step is spotting the thing that jams everyone up.

2

Move slow jobs out of the way

Hand the slow stuff to a helper working in the background, so nobody has to stand around waiting for it.

3

Keep the team moving

While that slow job runs in the background, the rest of your agents keep working without missing a beat.

4

Stop two agents doing the same job

When a thousand agents grab for the same task, exactly one gets it. No double work, no wasted tokens.

5

Lock it in so it can't break again

A built-in safety check stops the freeze from ever sneaking back. Once it's fixed, it stays fixed.

What changed

The 3 things this update actually did

If you want the proper name, the Hermes team calls it the "AsyncSessionDB" update (PR #54876). Here's what it did in three plain points — plus the proof they published.

01 · The problem

One slow job froze everyone

Every agent shared one "notebook". When one agent did a slow save, every other agent had to stop and wait. That's why big runs jammed up.

02 · The fix

Slow jobs go to a helper

Now the slow saves get handed to a background helper. Your agents never wait on each other — the team just keeps working.

03 · No double work

One job, one agent

If two agents reach for the same task, only one gets it. So you never pay twice for the same work, even with a huge fleet.

And they proved it with real numbers

0 → 105The old way got 0 jobs done during one slow save. The new way got all 105 — the team never stopped.
1000 → 11,000 agents grabbed for the same task. Exactly 1 got it. Zero duplicates, zero wasted spend.

Don't take my word for it — feel it. The demo right below is a live version you can break yourself.

See it yourself: your team's pulse

👉 Hit the red button and watch the green pulse flatline — that's your whole team frozen by one slow job, the old way. Then hit the green button and watch it keep beating — that's the team still working while the same slow job runs in the background, the new way.

Your AI team's pulse · try it live demo

See it: 1,000 agents, 1 job, no double-work

👉 Hit the button — a thousand agents all grab for the same job at once. Watch only one win it; the other 999 are politely told "already taken". That's how a big fleet never does the same work twice, and never wastes your tokens.

1000 Claims, 1 Winner live demo
The real use case

I ran a content factory on it — inside Agent OS

Demos are nice. A business result is better. So here's what this update actually lets you do: run a whole content team as a fleet of agents, on one board, without it breaking.

Inside Agent OS I flipped the Agent Kanban into Hermes cloud mode and gave it one topic. The team that ran it — a Planner, a Builder and a Reviewer, all Hermes agents, no local model. Here's the actual board mid-run:

The Agent Kanban running a Hermes content team — 3 articles done with previews, 1 building, shipping live to aimoneylabjuliangoldie.com
Real screen capture of my board: 3 articles already done (you can see the previews), 1 still being written, all shipping live to my site.

👉 Open the live board in Agent OS

Or watch it run right here — flip the toggle to "SEO cluster → Hermes", type a topic, and hit Assemble:

👉 Open the Agent Kanban in YOUR Agent OS

The flow, end to end

Planner (Hermes) broke my topic into five SEO article cards in 8 seconds.

Builders (Hermes) wrote every article — real 1,500–1,900 word posts following my actual blog-post SEO skill, keyword in the first and last line, FAQs, schema, the lot.

Each card claimed exactly once — the concurrency guarantee from the update, so no two agents wrote the same post.

Then one button deployed all five live to my funnel site. Saved to history. Indexable. Done.

This is the part the freeze fix unlocks: I'm not running one careful agent at a time anymore. I'm running a team, and the board doesn't fall over.

The five articles it shipped — live on my site

Every one of these was planned, written and deployed by the Hermes team in the Kanban. Click through — they're real, published pages:

All five written by Hermes agents, deployed to aimoneylabjuliangoldie.com, saved in your SEO history.

The shift

One agent at a time vs a fleet that holds

The old way

  • Run one agent at a time and babysit it.
  • Watch everything freeze on one slow job.
  • Find duplicate work where two agents grabbed the same task.
  • Cap your runs small so nothing locks up.
  • Burn tokens redoing what already got done.

The new way

  • Run a whole team of agents on one board.
  • Nothing freezes — slow jobs run quietly in the background.
  • Every task is claimed exactly once. Zero duplicates.
  • Scale the run up without the board falling over.
  • Spend tokens once, ship the whole batch live.
"But running a fleet of agents in Agent OS must cost a fortune in tokens."

It doesn't — and this run is the proof. The whole content factory ran on Hermes, not a paid metered API. You can run Agent OS on free local models, on free API tiers, and on the CLIs you already pay for — I plug my existing Claude subscription straight in at no extra cost.

And inside the AI Profit Boardroom there's a full set of token-efficiency tutorials, so you never have to think about this again. The freeze was the real cost. That's the part this update killed.

The full system

This was one board. Agent OS is the whole operation.

The content factory you just watched is a single surface inside Agent OS. Here's everything you get when you run the full operating system the way I do:

✓The Agent Kanban — fan one goal across a team of agents and ship the whole batch live.
✓The full Hermes agent fleet — Claude, Codex, GLM, Kimi, Grok and local models, one dashboard.
✓The 5-site SEO funnel — research, write, deploy and index real ranking pages.
✓Run it for free — local models, free tiers, and the CLIs you already pay for.
✓Token-efficiency training — so cost is never the thing that stops you.
✓3,600+ founders building alongside you, plus weekly coaching calls.
Get inside the Boardroom →
Where this goes

The cap on multi-agent work just came off

For a long time the honest answer to "can I just run a fleet of agents?" was "sort of, until it freezes." That answer changed with this update.

Slow jobs now run in the background. Your agents never stop to wait on each other. And two of them can't grab the same task. That's the unlock — and the content factory above is what it looks like in practice.

So pick a topic. Flip your Kanban to Hermes mode. Let the team run.

The freeze is gone. Go build the fleet.

Start today

Build your own agent factory this week

Everything in this guide — the Agent Kanban, the Hermes fleet, the SEO funnel, the token-efficiency playbook — is inside the AI Profit Boardroom, with the people and the coaching to get you running fast.

✓Step-by-step Agent OS setup — from zero to your first running fleet.
✓The exact Hermes + Kanban workflow from this guide, ready to copy.
✓5 live coaching calls a week to unblock you in real time.
✓1,000+ prebuilt AI agents and a community shipping daily.
Join the AI Profit Boardroom →

Or start free in AI Money Lab →

Built for operators · used in 38 countries · YouTube · X
Update: Hermes Agent · AsyncSessionDB · PR #54876 · NousResearch/hermes-agent