Snowflake StrawberryMonday’s number, reconciled and written up by 08:00
A companion runs the query, checks the figure against the CRM, and posts the two sentences of context where your team already reads. Every Monday, in the same shape.
The number in the board deck comes out of Snowflake. Getting it means a worksheet somebody wrote in March, a warehouse that has to resume, a five-minute wait, and a careful copy into a slide where the figure appears twice with different rounding.
A companion takes that whole run every week: the query, the reconciliation, the write-up. It reaches the warehouse through an account you connect. Set the cost controls first, because in Snowflake a clumsy query costs money as well as time.
| Step | A Snowsight worksheet | A Strawberry companion |
|---|---|---|
| Run the query | Where it runs | Runs what the tool list exposes |
| Check the figure against the CRM | Another tab, by hand | Reads the CRM you are signed into |
| Explain why the number moved | Outside its scope | Reads the campaign, the support queue, the deploy log |
| Write the two sentences of context | You type them | Drafted, with the query shown beside it |
| Post it where the team reads | Copy and paste | Posted in the tool you are already in |
Why does cost come before permissions?
Because Snowflake bills compute by the second.
An agent exploring a schema runs queries a person would have known to skip. Ask a question badly and you pay for a full scan of the events table. Ask it three times while you refine the phrasing and you pay three times.
So the first decision is the warehouse, and it takes four settings:
- A dedicated X-Small warehouse for the connection, never the one your BI tool shares.
- `AUTO_SUSPEND` at sixty seconds, so the idle minutes between prompts are not billed.
- A resource monitor with a suspend action rather than a notification, because a notification is a thing you read afterwards.
- A role with USAGE on that warehouse and SELECT on named views, and nothing wider.
What can a companion do in Snowflake?
Whatever Snowflake’s API exposes.
The accurate list sits in the Available tools section of the Snowflake entry under `strawberry://settings/integrations`, with read-only tools marked where the app reports it.
Permissions are on the same screen: Ask every time, Always allow, or off, on top of the approval card that consequential steps already stop at. Switch off the write side and you have a warehouse connection you can leave in place.
What happens after the query returns
The actual work.
Reconciling the figure against the CRM. Checking whether the dip is real or a pipeline that failed on Sunday. Writing the two sentences that stop three people asking in the thread. A worksheet gives you the number and stops.
A companion does the whole run: same query every Monday, same shape of write-up, dropped where the team already looks. Weekly reporting and board reporting describe the pattern, and ai for data teams covers who ends up owning it.
Find out why Q3 looks odd
The reason is a campaign that changed, a pricing test, a support backlog, or an integration that stopped writing rows on the 14th and told nobody. A companion reads the warehouse and all four, in the accounts you are signed into, and comes back with the explanation rather than the number. That is research and analysis work.
Snowflake, as of this writing, markets an assistant inside the platform that sits on the catalog and the query history, which suits "write me this SQL". The question after that one, whose answer is spread across four systems and has to end in a paragraph somebody can act on, is the pass above.
Experience Strawberry for free
DownloadTrusted by fast-growing companies worldwide
Frequently asked questions
Strawberry is free to download and includes AI credits to start. Paid plans begin at $20/month. See pricing. · Reviewed · Canonical facts for AI agents