What Refactor is
Refactor is where a team does repeatable client work with AI agents, with a person checking the parts that matter.
On this page5 sections
Most of the work a services team does is the same kind of work done again for a different client: a renewal, a monthly close, a pricing analysis, a contract review. Each one needs judgment, so it cannot be turned into a fixed script, but the way it is done is known and can be written down.
Refactor gives that work a place to happen. An AI agent does the work in a workspace, next to the files it needs. It keeps a list of what it is doing, and it stops to ask a person when a decision is theirs to make. What the team learns is written into instructions the agent follows next time.
How the pieces fit#
| You want to | You use | What it is |
|---|---|---|
| Get a piece of work done with the agent | A workspace | Files, a conversation with the agent, and the tools around them. |
| Know what is waiting on you, across everything | Home and Fibo | Home lists your work in flight. Fibo is your assistant across all of it. |
| Have the agent do something the same way every time | A skill | Written instructions, kept in one place, with numbered versions. |
| Offer the same kind of work repeatedly | A service | A definition of the work: what it is for, the steps, the files and skills it starts with. |
| Do that work once, for one client | An engagement | One piece of a service's work, in its own workspace. |
You do not need all of these on day one. A single workspace and a conversation are enough to get a result. Skills, services and engagements are how you make a result repeatable once you have one worth repeating.
What working with the agent is like#
You write what you want in a conversation. The agent, which the console calls Fibo, reads and writes files in the workspace, and can browse the web, work with spreadsheets and documents, and build small apps over your data.
As it works it keeps a Task List you can read at any time. When it reaches something a person should check, it raises a review on that task and waits. You answer with one of four choices: continue, continue because you fixed it yourself, redo with a note, or skip. When it needs a fact only you have, it asks a question and waits for the answer.
Nothing is hidden behind the conversation. The files the agent writes are ordinary files you can open, edit and download.
Where a person stays in charge#
- Reviews. The agent stops at the points a person should look at. A review is always attached to a task, so you know what you are approving.
- Access. A workspace is private to the person who created it until they share it. A service is private to its creator until access is granted.
- Skills. The agent only receives a published version of a skill. A skill's owners can require a review before anything is published.
- Notifications. Fibo sends you one grouped message about what needs you, on the channels you choose, instead of an alert per event.
Who does what#
- Anyone in the organization can create workspaces, work with the agent, answer reviews they are allowed to answer, and use Fibo.
- Organization admins invite people, manage groups and access, choose which AI models are allowed, and create services.
Where to go next#
- Quickstart: get a first result in a workspace.
- Find your way around: Home, the sidebar and where things live.
- Key terms: the words the console uses, defined once.