Buzz AI Agent: How Buzz’s Agents Actually Work (2026)
Let me break this down. A Buzz AI agent is a teammate that lives in your channels — tag it and it researches, writes, designs or builds, using the Claude Code or Codex subscription you already have.
From the default trio to custom agents and full teams, here’s how Buzz’s agents actually work.
Last updated: August 2026.
Key takeaways
Buzz ships with three default agents: Fizz (the maker), Honey (writing) and Bumble (research).
Create custom agents — a name, a job description, a provider and an emoji.
Group them into teams that coordinate autonomously — including a QC agent that checks the others’ work.
Each agent has its own keys, channel memberships and audit trail — siloed by design.
Out of the box, Buzz gives you three specialists: Fizz, the maker who builds things; Honey, who’s good at writing; and Bumble, the researcher. Add any of them to a channel, tag them with a task, and they get to work — with a live activity view showing exactly what they’re doing and a timer on their replies.
One rule to learn immediately: you must tag the agent every time. A plain reply does nothing — it’s deliberate, so agents don’t flood your channels.
Creating Custom Buzz Agents
The real power is rolling your own. A custom agent takes about a minute: name it, describe its job in plain English, pick its provider (Claude Code, Codex, Goose), and give it an emoji or image. Mine include an image designer (“design variations based on the image and example I give you”) and an SEO agent (“use your SEO skill to find keywords and publish content to WordPress”).
Because agents inherit your existing setup — Claude’s skills, APIs, even Search Console data and your Obsidian vault — a new agent is useful from its first message, no re-wiring required. You can even tell an agent to self-improve based on your feedback each time it’s used.
Agent Teams: Where It Gets Wild
Group agents into a team with shared instructions and tag the team instead of individuals. My marketing team — SEO agent, image designer, quality controller — researched keywords, drafted a post, generated images, then QC’d itself: the quality controller caught capitalisation errors and a wrong word count, the SEO agent fixed them, QC confirmed “ship ready”, and the post went live. One message from me, end to end.
Permissions, Memory and History
Capability
Why it matters
Own keys per agent
No shared credentials
Siloed channel access
An agent only sees what you allow
Audit trail
Every action is traceable
6 months of searchable history
Agents can find and reuse past work
Memory
‘Have we seen this bug before?’ works
Which Models Power Buzz Agents?
Agents run on the harness you pick — Claude Code was the most reliable in my testing, Codex works well, and Goose (Block’s own CLI) plugs in too. Hermes can join via the gateway — see my Hermes + Buzz guide — and you can even run free models behind it. For the full app tour, see my Buzz app guide and the complete walkthrough.
Get the Agent-Team Playbook
The full playbook — custom agents, teams, QC loops and the workflows that publish real work — is inside the AI Profit Boardroom Buzz masterclass.
An AI teammate in your Buzz channels — tag it and it researches, writes, designs or builds, powered by your existing Claude Code or Codex subscription.
What are Fizz, Honey and Bumble?
Buzz’s default agents: Fizz is the maker/builder, Honey specialises in writing, and Bumble handles research.
How do I create a custom Buzz agent?
New agent → name it, describe its job in plain English, pick a provider and an emoji. It inherits your existing Claude setup, so it’s useful immediately.
Can Buzz agents work as a team?
Yes — group them with shared instructions and tag the team. They coordinate autonomously, including QC agents that check and correct the others’ work.
Why isn’t my agent replying?
You probably didn’t tag it — agents only respond to tags, by design. On big tasks, give it time or nudge for the finished output.
The Bottom Line
Tag an agent, get real work back — that’s Buzz. Build proper teams with the masterclass in the AI Profit Boardroom.
