Module 1 — Getting Started · Lesson 1.6
The First-Week Checklist
The short version of the whole course, in the order that makes each step pay for the next
~12 min
What you'll learn
- Execute a full workspace setup in a defensible order
- Identify the four steps that everything else depends on
- Know what to deliberately defer until week two or later
- Tell whether the setup worked, using observable signals rather than a feeling
Kavanah rewards setup more than most tools, because the AI agent's usefulness is a direct function of how much true structure it can see. A workspace with real projects, real people, a connected calendar and a written charter produces an agent that is genuinely useful. A workspace with one project called 'Stuff' produces an agent that is a chatbot. This is the ordered path from the first to the second.
Day one: the four that everything depends on
One — set your timezone. Every scheduled thing in the product resolves against it, and fixing it later does not retroactively fix what already fired at the wrong hour.
Two — create your real projects. Not one catch-all. The project is the unit that access control, reporting, planning and the charter all hang off; work that lives in the wrong project is invisible to all four. Three to seven projects is a normal shape for a small team.
Three — get your people in and on the right projects. Membership plus per-project access. This is the step whose absence produces the 'I can't see anything' message on day two.
Four — connect your calendar. Of all the integrations, this is the one that changes the agent most immediately, because it is what turns 'schedule a call with them' from a suggestion into an action, and what makes capacity and availability mean anything.
If you do only these four, the workspace works. Everything below makes it better.
Day two to three: the surfaces your team will live in
Set up the board the way your team actually works. Custom statuses are the honest answer if your process is not To Do / In Progress / Done — a board whose columns lie about your process trains everyone to ignore it. Custom properties are for the fields you will filter and report on.
Connect email if you want the inbox and the agent's mail capabilities. Connect the tools your team already lives in — Slack, GitHub, the PM tool you are migrating from — so the work does not have to move all at once.
Start chat. A workspace where the conversation happens in Kavanah is a workspace where the agent can see the conversation, which is where captured work comes from.
Then leave it alone for a couple of days and let people use it. The next set of decisions is much easier to make with a week of real usage behind them.
End of week one: the charter and the agent
Now write the KVN charter, on the KVN page under Workspace. Know-how is what you can actually do; Vision is the end-state; Negation is the hard nos. It is worth doing now rather than on day one because you will write a better Negation after a week of watching what people actually ask for.
The Negation is the part that does real work. It is enforced: before the agent creates a task or a project, the work is checked against the charter, and work that crosses a Negation line is refused with the exact line that refused it, and logged. An empty Negation refuses nothing.
Then set the agent's autonomy. Start at safe — the agent may act, but only with internal, non-destructive tools; it cannot send email, post externally, or delete. Live with that for a while. Moving to full, where risky actions get parked for your approval, is a decision to make once you have watched it work rather than on faith.
What to defer, deliberately
Do not set up scheduled runs, exec meetings or the State of the Business report in week one. All three spend credits on a recurring cadence, and all three are much easier to configure well once you know what you actually want in them. They are off by default for exactly this reason.
Do not build automations before you have watched the manual version happen a few times. An automation encoding a process you have not yet settled is a process you now cannot change without remembering the automation exists.
Do not turn on task capture and then ignore the triage queue. Capture without triage produces a queue that grows until someone declares bankruptcy on it. Turn it on when someone has agreed to work the queue daily.
Do not invite the whole company on day one. Get five people working properly first; they become the ones who answer questions in week three.
How to tell it worked
Not by a feeling. Four observable signals, at the end of week two.
Is there work in the system that exists nowhere else? If every task in Kavanah is also in someone's notes app, the workspace is a mirror rather than a source of truth, and mirrors get abandoned.
Does the dashboard rail have something in it every morning? If it is empty, either nothing is happening or nothing is connected.
Has anyone asked the agent a question it answered from the workspace rather than from general knowledge? That is the moment the setup starts paying.
Does the KVN ledger show anything? A Negation with hits is a boundary that is doing work; one with none is either irrelevant or invisible.
Work the checklist
- 1
Three to seven, named after things your team would recognize. Not one called 'Work'.
- 2
Invite people and put them on projects
Membership in People; per-project access on each project's Team tab. Both are required — one without the other is the 'I can't see anything' bug.
- 3
Settings → Integrations. The single highest-value first connection, because it makes scheduling and capacity real.
- 4
Write the KVN charter at the end of the week
Know-how, Vision, Negation. The Negation is the enforced part, and an empty one refuses nothing.
- 5
Set agent autonomy to safe and leave it there for a fortnight
Settings → AI Agent. Watch what it does before you widen it.
What to watch
- Unique-to-Kavanah work
- The share of active tasks that exist here and nowhere else — not mirrored from a spreadsheet or a notes app.
- Healthy signal: Rising through week two. A workspace that stays a mirror gets abandoned when the mirroring becomes a chore.
- Members with a project
- Share of workspace members who have access to at least one project.
- Healthy signal: All of them. Anything less is someone staring at an empty app.
- First agent answer from workspace data
- Whether anyone has yet asked the agent something it answered by reading the workspace rather than from general knowledge.
- Healthy signal: Inside two weeks. It is the first moment the setup visibly pays, and it is what converts a sceptic.
Key takeaways
- ·Four things carry the setup: timezone, real projects, people on projects, a connected calendar.
- ·Write the charter at the END of week one — you will write a better Negation with a week of evidence.
- ·Start the agent at safe autonomy and widen it after you have watched it, not on faith.
- ·Defer anything that spends credits on a schedule: scheduled runs, exec meetings, State of the Business.
- ·Do not turn on task capture until somebody has agreed to work the triage queue.
That is the whole setup in one lesson. The rest of the course is each of those steps done properly — starting with the surface your team will spend the most time in: projects and the task board.