New framework · Hermes async subagents · 16 Jun 2026

The Hermes Agent Hive™.

New · Hermes Agent async subagents just dropped

Hermes just shipped asynchronous subagents. The delegate tool that spawns workers to fan out and do your work no longer blocks your chat. So I turned it into a hive: one Queen agent dispatches a swarm of worker agents that forage in the background — all at once — and each flies back with finished work. I pointed it at my SEO funnel and got 5 full articles from one keyword, written at the same time, hands-free. Here's the whole thing, live.

Hermes Agent · async subagentsdelegate_task(background=true)
0
dispatch time
then you're free
0
workers foraging
at the same time
0
turns you wait
nothing blocks you
Straight from the source

What actually shipped.

This isn't my idea — it's a real Hermes Agent release from Nous Research. Here's the announcement, and where to get it:

ASYNC BACKGROUND DELEGATION — dispatch in ~2ms · subagent runs on a daemon thread · result re-enters chat as a new turn · 13 + 137 tests pass

Get it + read it ↓

To turn it on yourself: run hermes update in your terminal. That's it — the delegate tool is now non-blocking.

What the hive built

One keyword. Five articles. At once.

This is the part I had to show first. I gave the hive ONE keyword — "how to build an AI agent" — and one command. It fired five worker agents in parallel, each writing a complete, distinct SEO article for a different site in the funnel. I didn't touch a thing while they ran. Here are all five — real output, 4628 words total, click any to read:

Step-by-step how-toA Beginner’s Step-by-Step Guide959 words · written in parallel · read it → Best-tools listicleBest Tools for Building Powerful AI Agents956 words · written in parallel · read it → No-code vs codeNo-Code vs Code-First Comparison1106 words · written in parallel · read it → Beginner FAQEssential Beginner FAQs on AI Automation792 words · written in parallel · read it → Agency case studyLessons from Our Real-World Agency Experience815 words · written in parallel · read it →

▶ five real articles · one keyword · written at the same time by background workers

Watch the hive run

The actual process, on tape.

No mock-up. This is the real terminal — the Queen fires the workers, they forage at the same time (watch the timestamps overlap at 07:27:32, four foraging at once), each returns to the comb, then a deploy worker ships the whole site. Real output, real timestamps.

▶ real run · five 6-second jobs done in ~14s (one-at-a-time would be 30s+) · then a live deploy

Why it never freezes your chat
old · synchronous — chat blocked the WHOLE task
background=true
2ms
↑ dispatched — then you're free the whole time the workers run →
The old delegate call blocked your chat until the task finished. background=true hands control back in ~2ms and the work runs behind you.
It doesn't just write — it ships

One command. A live site.

The hive doesn't stop at words. A background deploy worker ran netlify deploy --prod and pushed the whole funnel live — I didn't touch the keyboard between the prompt and the URL:

onekeyword THE HIVE5 articles · parallel deploy workernetlify deploy --prod 🌐live
🌐 LIVE → hermes-agent-hive-demo-jg.netlify.app

A real Netlify deploy on my account — a fresh demo site, so a live funnel is never touched. One keyword in → five articles and a deployed website out, hands-free.

My story · why this matters

I was you. Building my funnel one slow article at a time.

Before

My SEO funnel is five sites. Five articles per keyword.

I'd write one. Wait. Check it. Start the next. Wait again.

Every article blocked the one behind it.

A single keyword could eat half a day of babysitting.

Then the hive changed it.

After

Now I give one agent the keyword and walk away.

Five workers write five articles at the same time.

Nothing blocks me — I'm already onto the next keyword.

The whole funnel lands while I do something else.

You can have this too. Same tool. Same path.

The receipts

Real people. Real wins. Inside the Boardroom.

I'm not the only one running agents like this — agency owners, ecom founders, course creators, solo operators. Different businesses, same result.

2,200+ founders in AIPB
319k YouTube subs
$100k+/mo AIPB MRR
38 countries live

Real businesses, real wins, documented. Read the 158-page testimonials doc →

Before you scroll on —

Commit to building in parallel today. Not tomorrow.

You just saw five articles written at once from one keyword.

The next few minutes show exactly how the hive works.

So here's the deal.

Promise yourself one thing right now. You'll finish this guide and run hermes update before you sleep tonight. Just that one command. Because the moment your agents stop blocking you, the way you work changes for good.

The people doing it one-at-a-time are getting passed. The people running hives are the ones who look back in six months and say "that was the moment."

Commit to the transition. Commit to taking action today. This changes how much you ship.

The framework

The Hermes Agent Hive™.

A hive runs on one simple idea: the Queen doesn't do the work — she sends out workers, and they all forage at once. Here's how the five parts give you a funnel that builds itself.

background · daemon threads QUEEN your lead A1 A2 A3 A4 A5 5 workers · forage at once THE COMB results re-enter → your chat
i.

The Queen

Your lead agent. You hand it one goal — "run the funnel for this keyword" — and it decides the fan-out. It never does the grunt work itself.

ii.

The Workers

One background subagent per job, each dispatched in ~2ms with delegate_task(background=true). Each runs on its own daemon thread with its own goal and context.

iii.

The Forage

They all work at the same time, off to the side of your chat. Dispatch returns instantly — you're never blocked. Keep planning, keep chatting, start the next keyword.

iv.

The Comb

As each worker finishes, its result flies back and re-enters your chat as a new turn — self-contained, in order, no mid-loop mess. You collect finished work as it lands.

v.

The Hive Mind

It all lives in your Agent OS — one memory, one dashboard. Every worker knows your business, so the whole hive compounds instead of starting cold.

The comb · each worker returns on its own, as a new turn
worker-2
↩ new turn
worker-3
↩ new turn
worker-1
↩ new turn
worker-4
↩ new turn
worker-5
↩ new turn
They don't finish together — each result flies back the moment it's done, re-entering your chat in order (these are the real return times from the run above).
Old way vs new way

Two ways to build a five-article funnel.

One at a time — each article waits for the last
A1A2A3A4A5a long morning
The Hive — all five start at once
A1
A2
A3
A4
A5minutes
Same five articles. Serial, they queue behind each other. In the hive, they all start at the left edge — the wall-clock is the slowest single one, not the sum.
Old way · serial
~ half a day
  • Write article 1, sit and wait for it
  • Check it, then start article 2
  • Each article blocks the next in line
  • You babysit every single step
  • The chat is frozen while it works
  • Result: one keyword eats your morning
New way · the Hive
~ minutes, hands-free
  • One command: "run the funnel for this keyword"
  • Five workers fire at once (~2ms each)
  • They write all five articles in parallel
  • You walk away — nothing blocks you
  • Each finished article flies back on its own
  • Result: the whole funnel lands while you do other work
How to run it

It's one command away.

I built you a ready-made hive. The seo-lead profile is the Queen (the delegate toolset, five concurrent workers, auto-approve so they write unattended), and the seo-funnel skill is the playbook. Turn it on and point it at a keyword:

# 1. switch the delegate tool to non-blocking (one time) hermes update # 2. fire the whole funnel from one keyword hermes --profile seo-lead -z "Run the SEO funnel for the keyword 'your keyword' across 5 sites"

Five workers fan out, five articles land. The same shape works for any fan-out — it doesn't have to be SEO.

Get the whole hive in one place

Want the Hive wired in for you?

You just watched one keyword become five articles, hands-free. Now imagine that running next to every other model and agent you use, on one screen, sharing one memory — and pointed at your whole business, not just SEO.

That's the Agent Operating System inside the AI Profit Boardroom. Hermes, Grok Build, Claude, GLM, Kimi — one dashboard, one memory, builds previewing live. The hive plugs straight in, with your business context baked in.

Get the Agent OS → Inside the AI Profit Boardroom · link in the description
Not just SEO

Five more things to fan out.

The hive is a pattern, not a one-trick tool. Anything you'd do one-at-a-time, you can now fire in parallel and walk away:

Model bake-offs

Build it on every model at once

Fan out the same build to Grok, GLM, Kimi and Qwen as workers — compare the results as they land instead of running them one by one.

Content pipeline

One transcript → ten outputs

From a livestream: workers write the description, the Substack post, the Shorts scripts and the thumbnail brief — all at once.

Overnight batches

Queue it before bed

Fire ten jobs at midnight; they forage on daemon threads and the finished work piles up as new turns by morning.

Research while you build

Never stop to look things up

Send a worker to research in the background while the Queen keeps coding; the findings fly back when ready.

Should you turn it on?

The three things holding people back.

If I'm not watching each step, it'll go off the rails.

Each worker gets one tight goal and its own context, and every result flies back for you to check. You're the Queen — you set the jobs, you keep the honey.

Running agents in parallel sounds like chaos.

It rides the same rails as background terminal jobs — message order and prompt-cache stay intact. 13 targeted and 137 delegate tests pass. It's boring and safe by design.

This is only for coders.

It's one command and a plain-English goal. If you can type "run the funnel for this keyword," you can run a hive.

Don't take my word for it

158 pages of members who already broke through these exact beliefs — real businesses, real wins.

Read the 158-page testimonials doc →
The recap

What you walk away with.

i.

You stopped waiting. Dispatch is ~2ms — the work runs while you keep going.

ii.

You stopped babysitting. One command fans out the whole funnel.

iii.

You ship in parallel. Five articles at once, not one after another.

iv.

You collect, you don't chase. Each finished job flies back to you.

v.

It's not just SEO. Bake-offs, content, overnight batches — same hive.

vi.

It's one command. hermes update, then point a Queen at a goal.

Build the new way

Run a hive every day.

If you want your agents to actually multiply your output — not just be one more tool you babysit — grab the Agent Operating System inside the AI Profit Boardroom.

It turns Hermes, Grok Build, Claude, GLM and Kimi into one system with shared memory and one dashboard. Every new feature — like async subagents — makes the whole hive more powerful, automatically.

Get the Agent OS → Inside the AI Profit Boardroom · link in the description

Run hermes update first. Point a Queen at a keyword second. I'll see you in the next one.