Rolling Kavanah Out to a Team

Week one: five people, real work

Do not invite the whole company.

Pick five people doing work that is genuinely coordinated, and set up their projects properly — real structure, real access, calendar connected. Their work moves in, entirely. Not a pilot project, not a trial: the actual work.

The reason for five rather than fifty is that week three depends on having people who can answer questions, and you only get those by letting a small group become genuinely fluent first.

By the end of the week, those five should have work in Kavanah that exists nowhere else. That is the test.

Week two: make it personally better

The first two weeks of any rollout are cost for the participants. Week two is where you pay some of it back.

Get everyone's daily digest scheduled. It is the most reliably appreciated thing in the product for an individual, because it arrives before they start and removes a decision.

Get them using the agent for something tedious — summarizing a thread, drafting a reply, breaking a brief into tasks. One useful delegation each is enough.

And fix whatever annoyed them in week one. Board columns that do not match how they work, notifications that are too loud. These are small and they are the difference between 'this is being done to me' and 'this is mine'.

Week three: widen, and use your fluent people

Now invite the rest, in groups rather than all at once, and route their questions to the five people from week one rather than to you.

That routing is the whole point of the sequencing. A question answered by a peer who was in the same position a fortnight ago lands very differently from one answered by the person who chose the tool.

Write the onboarding playbook this week, in Knowledge, while the questions are still arriving. It will be much better than one written in advance, because it will answer the questions people actually have.

Week four: the charter and the agent

Now write the KVN charter, with input from the people using the workspace. You have a month of evidence about what gets proposed and what should not.

Set the agent's autonomy ceiling to safe and let people find their own level under it.

And do the first weekly review in public — walk through bottlenecks, staleness and clustering with the team watching. That demonstrates what the data is for, which is the thing most rollouts never do and which is why most teams' tools stay input-only.

The four objections, and what answers them

'This is more admin.' It is, for two weeks. What answers it is not argument, it is the digest and one useful agent delegation. Something has to become easier for them specifically.

'We already have a tool for that.' The honest answer is that Kavanah's value is having them in one place where an agent can see across them, and that connecting rather than migrating is available. Do not claim it is better at the thing the other tool does best.

'I don't trust the AI.' Set autonomy to assess for a fortnight and let them watch. Evidence beats reassurance, and this is what that setting exists for.

'Nobody else is using it.' This is the real one and it is a sequencing problem. It is answered by having the five people from week one visibly working there, which is why they came first.

The conversion moment

For almost everyone, adoption turns on a single moment: they ask the agent something and get an answer that could only have come from their own workspace.

Not a generically good answer. A specific one — what is overdue, who is overloaded, what the client agreed in March.

Engineer that moment. It requires the setup to be real, which is why the whole course is ordered the way it is, and it usually requires someone to prompt them to try. In week two, ask each person to ask the agent one question about their own work.

After that moment, people learn the rest themselves. Before it, they are complying.