The short version: install it, say who is responsible for what, take the applications most companies start from, load your people, then build one of your own and watch it happen.
A day, not a quarter.
1. Get it running
In the cloud, if you would rather not run anything: register and you are in.
On your own infrastructure, if the data has to stay inside: the same image, brought up with compose. No cloud licence server to call home to, and it runs in a closed network. See Deployment.
Nothing about the two differs afterwards, so this is a decision about where the data lives, not about which product you get.
2. Say who is responsible for what
Before anything gets built, decide who owns which part of the platform. A small set of functional roles maps onto the way companies already divide this up:
| Role | Owns |
|---|---|
| Tenant administrator | The organisation itself: people, access, what is installed |
| Content manager | Applications and pages - authoring, publishing what your company runs |
| Data manager | Sources, datasets and models - what the numbers mean |
| Process manager | Processes, routes, automations |
| AI manager | The AI employees: their scope, their skills, their budgets |
Two things worth doing in the first hour: name a person for each, and give everybody else the ordinary access that lets them use what is built without changing it.
See Roles and access.
3. Install the applications most companies start from
You do not have to build anything to have something working today. Four of them are a foundation almost every organisation needs:
- Task management - boards, owners, deadlines, checklists. The layer that is otherwise a chat and a spreadsheet.
- Corporate portal - the place people open in the morning: news, documents, the forms they need, their own tasks.
- Project management - what is running, who is on it, what is late.
- Product and catalogue management - the lists everything else refers to.
Install the demo profile first if you want to see them with data in them before anybody types anything real. Then install them empty for actual use.
See Solutions.
4. Load your people
Everything routes through your organisation - who approves, who stands in when somebody is away, who sees which rows - so this is the step that makes the rest correct.
Today: an export. Take the structure out of your HR or HCM system as a CSV - units, positions, employees, who reports to whom - and load it. Load the same structure again and it converges, so keeping it current is a repeated step and not a clean-up.
From that moment a route saying "the head of the requester's department" resolves to a real person, and keeps resolving correctly when that person changes department.
Soon: a direct connection. Connectors to the usual HCM systems - BambooHR and its neighbours - are close, and they will keep the structure in step without anybody exporting anything. The CSV is the path today, and it is not a dead end: what it loads is the same structure the connector will keep updated.
See Org structure.
5. Build your first application
Pick something small and annoying. A good first one, because every company has it and nobody has a system for it: idea management - somewhere people put suggestions, somebody reviews them, the good ones become work and the rest do not vanish.
Describe it:
People submit an idea with a title, a description and which area it affects. It goes to the head of that area. They can accept it, decline it with a reason, or ask for more. Accepted ideas become tasks with an owner. Everybody can see the board and comment. Show me how many were submitted and accepted per quarter.
What happens next takes minutes, not weeks:
- The assistant asks the questions you did not answer - what if the head is away, can an idea be resubmitted, who sees declined ones - and writes a specification.
- You read it and approve it. This is the moment worth spending ten minutes on.
- The agent builds it: the record and its fields, the statuses and the route, the roles, the board, the dashboard, the page.
- It runs on your data. Open it, submit a test idea, walk it through the route, look at the numbers.
- Move things on the canvas, or tell the agent what to change. Repeat until it is right.
See Built by AI, approved by you.
6. Hire your first AI employee
The application works. Now give it a colleague.
Stay with the idea board, because it shows the shape without needing a big process. Describe the job the way you would to a new hire:
When somebody submits an idea, work it up. They will write three lines; turn that into something a reviewer can actually decide on - what it is for, the problem it solves, who benefits and by how much, what it would take, what it touches. Ask the author if something essential is missing, and show them the write-up before it goes to review. Watch the board: if an idea has been waiting more than five working days, remind whoever it is with, in Slack or Telegram if that is where they are. When a new idea looks like one submitted before, link them and say so in a comment. If an idea mentions a customer complaint or a safety issue, tell me straight away rather than waiting for the queue.
The first sentence is the one that matters. That is the analyst's job: taking a rough thought and turning it into a statement somebody can decide on - the goal, the problem, the value, the effort, the scope.
It is also where most ideas die. The person who had the idea is rarely the person who enjoys writing it up, so they put down three lines, and three lines is not enough to decide on, so it waits, and then it is stale. An organisation loses its good ideas in the gap between having one and writing it down properly, not at the review stage.
Then set the boundaries, which is the part worth thinking about for a minute:
- It may read every idea, work it up into a proper brief, ask the author clarifying questions, comment, link duplicates and correct the area an idea was filed under.
- The author sees the write-up first and can correct it. It is their idea written up, not replaced by somebody else's version of it.
- It may not accept or decline anything. That is the reviewer's decision and it stays that way.
- It may not delete, and it may not write to anything outside this application.
- Everything it does appears in the record's history under its own name, next to what people did.
Within a week the difference is specific. Ideas arrive written up instead of gestured at, so the reviewer decides instead of asking for more. Nothing sits in the queue unnoticed. And people submit more, because writing three lines is no longer the price of being taken seriously.
And it is part of the application. The agent is declared alongside the processes and the pages, so it installs with them, versions with them, and travels when the app is copied to another division, instead of being a configuration somebody made once and nobody can find.
Start with one, and with a job small enough that you can tell whether it did it well. Widening its scope later is an edit. See AI organisation and AI employees.
7. Publish it, and let people use it
Publish the version and let it out to real users - a team first, not the whole company.
Then collect what they say. Comments sit on the records themselves, so the feedback arrives attached to the thing it is about, not in a thread somebody has to summarise. Watch where people get stuck, what they leave empty, which status everything piles up in.
And change it. That is the part that matters more than anything above: a week later you edit the description, publish a new version, and it is different on Monday. The people using the old one are not disturbed until it is installed.
What you have at the end of the day
A platform running where you chose. An organisation with owners for each part. Four applications your teams can use immediately. Your real structure inside them. One application you built yourself, in front of the people it was built for. And an AI employee working in it, inside limits you set.
That is enough to judge the rest of it honestly.
Next
What you can do with it - a wider set of ideas for what to build second, and How it works for what the platform is doing underneath the six steps above.