Grow your staff
Say "I want someone who just does our bookkeeping", and your main assistant asks a few questions, puts a shortlist of ready-made specialists on screen beside the chat, and builds the one you pick. A minute later there is a new card on Manage > Agents (/agents), and a chat waiting where your new bookkeeper asks how you like the books kept.
That is how Player One grows. You do not configure an agent: you describe the job, and someone turns up to do it. The catalogue and the build are the Forge.
The team you start with
A new installation is deliberately small. You get your main assistant, the one that knows you and takes everything first. Behind it sits a general worker for routine side jobs your assistant would rather hand off, and a coordinator, which plans work too big for one agent and shares it out but never does the work itself. A couple of system agents keep your memory tidy in the background and take no ordinary instructions. Everyone else, you add when you want them.
Asking for someone new
You ask in plain words. "Make me an SEO specialist." "Build a researcher who only works on the Hanley account."
Your assistant settles the shape with you in one conversation, then builds it. The new agent gets:
- A role. What it does and what it is good at. The coordinator reads exactly this when deciding who to hand a task to.
- A workspace of its own. Its own place for files, so its drafts stay out of your way.
- A memory of its own. It follows the agent wherever it works, so what it learns stays with the agent doing the job.
- A bounded set of tools. Not everything Player One can do: the things this role needs. The list is on the agent's own page, under Advanced.
- A face and a name. Ready-made specialists arrive with theirs. For one built from scratch, your assistant shows you a few faces and you choose.
Only your main assistant can create, change or retire an agent. An agent cannot widen its own permissions or make itself a colleague.
The ready-made specialists
You rarely start from nothing. Player One ships a catalogue of 77 ready-made specialists grouped by trade: a bookkeeper, a recruitment specialist, a technical writer, a market researcher, an SEO specialist, a customer service agent and many more.
When you describe what you need, your assistant searches that catalogue and puts the closest few on screen as tiles in the preview pane beside the chat. You look through them, ask about one, and say go when you are happy. Nothing is built until you say so. If nothing fits, say so and your assistant writes the role itself.
Why a specialist beats a generalist
An agent that does one kind of work gets better at it, because its memory is its own: every correction you make lands in one place rather than being diluted across everything you have ever asked for, and the skills it uses collect notes from its own runs. Six months of bookkeeping makes your bookkeeper genuinely good at your books, in a way that six months of everything makes nobody good at anything.
Agents, delegation and team chat covers when to hand work over and when to keep it in front of you.
Who works for whom
Once you have more than a few, you tell Player One how they fit together: "Nadia runs outreach, Bob does the books, both work to my assistant." Your assistant writes that down as a team page: who leads, who mainly does what, then a roster of each agent and one line on what it handles. Every agent on your team can read that page and knows where it belongs, so when work arrives at the wrong desk it knows where to send it.
When the roster outgrows one page you split it: Operations moves to a page of its own with a manager, and the top page becomes an index of teams. They sit with your other shared skills on Admin > Skills (/admin/skills).
An assistant you can put in front of the public
"Make a support assistant our customers can message" builds something different: a sandboxed agent, standalone by design.
It knows only what you gave it. Write it a role and a few pages of knowledge, your refund policy and your shipping FAQ, and that is its world. It cannot open your vault, browse your files, search the web, read your other conversations or talk to your other agents, and it cannot change any of that, including for itself. It is not on the team either: no jobs, no delegation.
Then you give it exactly the actions you want it to have; Build with it covers how such a tool is made. "Let it look up an order" becomes one narrow tool: it asks for the order number and the email address, it gets back the status and the delivery date, and nothing else comes through. Your credentials stay on your side of that wall, where the agent never sees them.
Five of the ready-made specialists are built for this work: a website concierge, a booking assistant, a tier-one support agent, an order status tracker and a community greeter. Customers reach it through a channel you connect, such as WhatsApp or Telegram; its own chat inside Player One is where you test it. It leaves you notes on its Workspace tab about what it is being asked, and what it writes there never returns to its own prompt.
Create a support agent for your customers walks it through, and Vault and sandboxing covers where your passwords sit while it works.
Where this lives on screen
Manage > Agents (/agents) is the gallery: a card each, filed under tags you arrange yourself, with New agent to open the Forge, though asking your assistant is the usual way. A card opens that agent's own page (/agents/<name>), holding its profile, skills, workspace and settings, with its tools and memory under Advanced.
Manage > Active Agents (/manage/active-agents/by-project) shows who is working right now, and Manage Project > Agents (/project/agents) decides who belongs to which project.
Where next
- Agents, the short version of what an agent is and how it fits.
- Give it work, how a job spreads across the team.
- A support bot on WhatsApp with your knowledge, one worked example end to end.