Cognition, the maker of Devin, connected it to its own databases so staff can ask data questions in Slack and get an answer plus the SQL behind it.
- First reported
- devin.ai · 1 Aug 2025
- Added
- 13 Oct 2026
- Source checked
- 13 Oct 2026
What they did
The team linked Devin to its warehouse and tools such as Neon and Sentry through MCP servers, so it can read schemas and run queries without seeing database passwords. Devin also searches the codebase to learn how each metric is calculated. Staff ask things like conversion rates by industry and company size, feature adoption after a launch, or next quarter's user growth forecast.
What happened
The company says this kind of question now comes up dozens of times a day, and estimates it would cost a data team more than 150 hours a month to answer by hand. Each Slack reply includes the final query so the asker can check and rerun it.
Keep in mind
This is Devin's maker describing its own use, and the 150-hour figure is an estimate that assumes about an hour per question. The post gives no accuracy rate and lists no failures.
Try it with your agent
You have read-only access to <your database or warehouse> and can read <your codebase or metric definitions>. When someone asks a data question in <Slack channel>, find the right tables, check how the metric is defined, run the query and reply with the answer, a short explanation and the exact SQL. If the data has a gap or the definition is unclear, say so instead of guessing. Never write to or change any table.
Replace the <angle brackets> with your own details.
Original source
devin.aiVendor case studyPublished by the company that makes the agent. Read it as a claim.
Tools for this kind of work
- RepoExtract tables and text from complex or scanned PDFs with Docling
- RepoSearch the web and scrape sites into clean data with Firecrawl
