Most teams draw a hard line between warehouse data and spreadsheet data. The AI doesn't need that line. The harness does.
Most companies run two data systems simultaneously. There's the formal one — Snowflake, BigQuery, a data warehouse with a schema and a governance team and a ticket process for access. And then there's the real one: a collection of Google Sheets that the ops team maintains, that the finance team relies on for close, that everyone uses but no one wants to admit is load-bearing.
Both are real. Both have data that matters. And for most AI tools, you have to pick one.
The problem with picking a lane
If your AI assistant only sees the warehouse, it gives you clean answers to clean questions. But your actual work lives in the gap between the warehouse and the tracker. The Snowflake pipeline says one thing; the spreadsheet the regional lead maintains says another. Figuring out why is usually a manual exercise — someone pulling data from both, pasting into a third sheet, squinting.
If the AI only sees the spreadsheet, it's reasoning over whatever the ops team typed in last Tuesday. Useful, sometimes. Trustworthy as your source of record, no.
The line between warehouse data and spreadsheet data is an organizational habit. Teams drew it because their tools required it. AI doesn't require it — but most AI tools inherited the same constraint and never questioned it.
What a unified context actually enables
When an AI harness can hold both sources at once, the questions you can ask change. Not just "what does Snowflake show for Q1 revenue?" but "how does the Snowflake figure compare to what the finance team has in the close tracker, and where do they diverge?" That's not a complex question. It's just one that previously required two tabs and a person in the middle.
The AI doesn't need the data to come from one place. It needs context — structured, current, scoped to what the user is actually allowed to see. The harness's job is to provide that context regardless of where it originated. Snowflake and Sheets are both just sources. The model treats them the same way.
The hard part is still the plumbing
Connecting to Snowflake is straightforward. Connecting to a Google Sheet is straightforward. Keeping both fresh, normalizing them into something a model can reason over coherently, enforcing access rules that span both systems, and routing the output somewhere useful — that's the work. It's not glamorous, but it's what determines whether the AI gives you an answer you can act on or one you have to go verify manually anyway.
The irony is that the harder you work on the data layer, the simpler the AI interaction becomes. A well-structured harness turns a complex cross-source question into something that answers in seconds. A poorly structured one makes even simple questions unreliable.
What this looks like at Brooked
Brooked connects to both Snowflake and Google Sheets and treats them as a unified surface for AI queries. You can ask questions that span your warehouse and your working documents — get answers that account for both — without manually reconciling them first. The model gets context from wherever the relevant data lives; you get an answer.
That's the point. The data's location is a detail the AI shouldn't have to care about. And with the right harness, it doesn't.