The Data Team Stack

The Switchy Project a small data team works through. Sonnet 4.7 as the default for writing and explaining SQL, drafting transformation logic, and turning a vague stakeholder ask into a precise, answerable question; Opus 4.7 for the gnarly reasoning — debugging a metric that drifted, reconciling two numbers that should match and don't, designing a model nobody will have to rebuild in six months; DeepSeek R1 when you want a cheaper long-chain reasoning pass on a thorny derivation and a second take on the math. GitHub puts the dbt repo and the pipeline code in the chat: "@github what changed in the staging models last deploy", "@github explain the diff on this transformation PR". Supabase reaches the data itself for a quick gut-check: "@supabase how many rows landed in events yesterday". Notion is where definitions live so a metric means one thing: "@notion what's our canonical definition of active team". Slack is how the answer gets back to the people who asked: "@slack draft the update for #data-requests on the churn question." The payoff for a data team specifically is shared, durable context about *what your numbers mean*. The Project remembers the definition of "activated team," the caveats on that one flaky source, the reason a metric was restated last quarter — and exposes that memory as an MCP server, so it shows up inside every analyst's Claude Code, Cursor, Claude Desktop, and ChatGPT. Less re-litigating "wait, how do we define that again" in three different threads. Start free on Starter, invite the rest of the team when you're ready — the context comes with them.