Your first week
Week one has one theme: make the drafts good enough that loosening autonomy is a small decision rather than a leap. That means adding knowledge where drafts were vague, writing standing instructions where you kept editing the same sentence, turning on one more channel, and letting the first follow-ups come round so you can see the chase actually happen.
Let the edits tell you what is missing#
The most reliable guide to what your Knowledge lacks is your own editing. Every time you rewrite the same part of a draft, that correction belongs somewhere durable. Which somewhere depends on how widely it should apply:
| The correction | Put it in | Scope it reaches |
|---|---|---|
| A fact about your business is wrong or missing | Knowledge | Every answer, every channel |
| The tone is not yours | A standing instruction on Connect Rules | Future behaviour, stated once |
| Something specific to one customer | Memory at the contact tier | That person only |
| Something true of one mailbox or number | Memory at the endpoint tier | That mailbox or line only |
| This person should not be contacted at all | A block directive on the contact | Every channel, permanently, until removed |
Add one channel, not three#
A second channel doubles what you have to judge. Add the one your customers actually use — for many businesses that is WhatsApp, for others the phone — and give it the same treatment email got: hold everything, read what is written, then decide.
Configure the channel's provider on
#/integrations, or on that channel's own screen where it has one.Result Inbound arrives and appears alongside email in the same conversation list, resolved to the same people.
Set that channel's autonomy explicitly on
#/autonomyrather than relying on the workspace rule.Result The channel-scoped rule now beats the workspace one, so a change to email later cannot quietly change WhatsApp.
Watch one full day of it before touching the mode again.
Result You have seen how the channel's shorter, faster messages differ from mail — which is the whole reason to scope autonomy per channel.
Let a follow-up come round#
Follow-ups are the part that is hard to evaluate in an afternoon, because the interesting moment is when one falls due. Create a couple deliberately in the first days — a chase on a quote, a check-in after a delivery — and watch what happens at the due time.
- Follow-ups exist on
email,whatsapp,sms,phone,anyandtask. - The
taskchannel is work for a person. Nothing ever sends it; it appears as work to do, which is exactly right for 'ring the supplier yourself'. - Phone follow-ups drain a couple at a time and go through the same gates as any outbound call — a line that is not ready is retried later rather than forced.
- A phone follow-up more than a day late is closed as missed rather than rung, because a very late call-back is worse than none.
Deciding whether to loosen autonomy#
By the end of the week you have a body of evidence: the drafts you approved unchanged, the ones you edited, and the ones you rejected. Loosen narrowly rather than globally. Autonomy resolves contact → endpoint → channel → workspace with the narrowest winning, so a sensible first move is autonomous on one quiet endpoint — a mailbox handling order confirmations, say — while the rest of the workspace keeps asking.
If the answer is 'not yet', that is a legitimate long-term setting. Some workspaces run on ask_before_send permanently and get most of the value: the writing, the research, the memory and the chasing all still happen — only the final send waits for a person.
Questions#
How much knowledge is enough?
Enough that a draft stops being generic, which is usually less than people expect. Grounding material is budgeted before it reaches the model, so adding a fifty-page policy document does not put fifty pages into every answer. Short, specific, correct entries beat volume.
Should I write rules or just correct drafts as they come?
Both, in that order of preference. Correcting a draft fixes one message; a standing instruction changes the behaviour. The signal that you need a rule is repetition — the third time you make the same edit, stop editing and write it down instead.
Is a week the right amount of time?
A week is a convenient unit, not a threshold. What actually decides readiness is volume: a workspace that saw two hundred messages knows more about its own drafts than one that saw six, whatever the calendar says.