The workshop
One day with your developers and DBAs, built around your own work
I've spent a long time teaching people to tune queries and indexes by hand. Nobody wants to learn that way anymore, and I don't blame them. These days I let AI do the annoying stuff, and I spend my time deciding what's worth doing. That's what this workshop teaches your team.
The bigger change is how they think about the job. They stop thinking of themselves as their job titles, and start thinking of themselves as product managers for their jobs. The database is the product, and they manage the agents to make it the best one they can. They decide what the agents work on, when it's good enough, and what ships.
Yes, AI gets things wrong. So do people. Ask someone to write a DMV query from memory and they'll screw up a bunch of stuff too. Let them run it, read the errors, and fix them, and things work out in the end. AI is the same way. The more you let it get into things, the better it gets at catching and fixing its own mistakes.
It's delivered live, on-site or remote: about four hours of core material, plus extension tracks that fill out the day, up to eight hours. It's built around your team's actual queries, plans, and schema. I demo with Claude Code, Performance Monitor, Performance Studio, and my query plan plugin, and the labs run in whatever AI tools your team already has.
Chat window vs. agent: the same model guesses in a chat window, and gets there when it can run the query, read the error, and fix it
Make it a copy of you: a login with the rights you'd give a new DBA, an instructions file and skills for how you work, and permission rules and hooks for what it should never, ever run
Your monitoring data through MCP: real waits and query history, and the difference between "nothing collected" and "nothing happened"
Plans without the squinting: tools built around the ways models misread execution plans
Reproduce it: the parameter values that hurt and production's plan, on a copy you can break
Let the robot do permutations: a dozen rewrites and index options, every result checked, every timing in a table. You pick.
Lab, your slowest query: thirty seconds to ten seconds. Cool! Where are those ten seconds going?
Where it still goes wrong, and what catches it
Picking the right model for each job, and measuring what each accepted change costs, so the bill doesn't grow faster than the results
Before the workshop, your team fills out a questionnaire: who's coming, which AI tools you're allowed to use, which servers you run, and one to three of your slow queries. Your slow queries replace my StackOverflow examples wherever you let me. Then pick an extension track:
Tuning Lab: more of your slow queries through the loop, in pairs
Build Your Copy: write the instructions file, skills, permissions, and hooks for your own environment
Review Before It Ships: point an agent at procedures and migrations from your own pull requests
I've worked with 600+ SQL Server clients. I use this setup every day: I build Performance Monitor, Performance Studio, and a set of Claude Code plugins with AI agents, under rules I wrote to keep them correct, safe, and cheap. That's the tooling your team uses in the workshop.
Live instruction (on-site or remote via Teams/Zoom)
The pre-workshop questionnaire, and a day set up around your answers
Exercises on your own queries, plans, and schema
Everything from the day: the scripts, the sample instructions file, the skill, the permission rules, and the hook
30 days of email support after the workshop: answers within two business days, plus one review of your team's instructions file and permission rules