About
AI for DBAs is the work of one career-long database administrator who got tired of writing reports nobody read.
Who's Ward
I'm Ward Minson. I have been a SQL Server DBA for twenty years.
The middle of that career was the regulated end of the spectrum: PHI under HIPAA as a Principal DBA at Lumeris, then four years of NERC CIP work for Entergy through ComTec, where I built the SSRS reporting that carried us to 100% compliance in NERC CIP audits. The job in those environments is variance detection. You need to know when a database is behaving differently than it was last Tuesday, and whether the difference matters. The job is also paperwork, because the data and the schema are the kind of information that legally cannot leave the network without a vendor agreement.
I am currently the Senior SQL Server DBA at Cloudingo, where I run de-duplication pipelines for enterprise Salesforce customers under GDPR, PCI DSS, and SOC 2. I am also a Navy veteran, which mostly explains my tolerance for being awake at 3 AM holding a flashlight.
Across all of those years one thing has stayed constant: the data was always there. The interpretation was always the bottleneck. PowerShell automation cut maintenance overhead by 70%, and AI assistants like GitHub Copilot, GPT-4, Claude, and Gemini cut the analysis bottleneck dramatically further. But none of that worked in the production data path of a regulated environment without a different kind of architecture.
Why Bob exists
Bob is the autonomous SQL Server agent I started building in 2024 to close that interpretation gap. Eighteen months later it is the system documented in The Birth of Bob: a three-layer agent that runs entirely on a single RTX 3090 in my home lab.
- Sensors probe SQL Server DMVs on a 60-second cycle.
- Reasoning runs on a local Ollama instance with
gemma4:26b, never sending a query plan to an external API. - Actuators are gated by a snapshot-before-apply discipline that has caught more than one bad fix before it landed in production.
Bob exists because the conventional answer does not scale. That answer is having a senior DBA read the 214-page health report manually every Monday. It costs money and concentrates rare interpretive expertise in too few people. Cloud LLMs were ruled out for the same regulated environments where the report problem hurts most. So Bob runs locally. Query plans never leave the LAN. The audit answer is one sentence: "We process this data on our own hardware using open-weight models."
What you'll find here
This site is the field notebook for that work: what's working, what isn't, and what I have learned along the way that's worth sharing with other DBAs walking the same road. Three places to start:
- The book: The Birth of Bob. 69 pages, free to read, free to download.
- The notes: the blog. Short pieces on AI for DBAs in practice.
- The newsletter: AI for DBAs on LinkedIn. Weekly editions.
If you are a DBA working in a regulated environment and the report problem feels familiar, you are probably my reader. The form below is the easiest way to reach me. It lands in my inbox the moment you hit send.