Every data team is experimenting with AI. Far fewer have turned it into workflows that produce the same result twice. Ask the same question on two different days and you get two different answers. Each engineer keeps their own prompts. Write access gets handed out like read access. The answer isn’t less AI. It’s structure.
This three-part series walks through the three levels of agentic data engineering:
- Ad hoc prompts that speed up daily work
- Standardized skills your team builds so a task runs the same way every time
- Production playbooks and built-in agents to support end-to-end data workflows with guardrails
The patterns apply to any data team working with MCPs and AI agents, whether or not you run Coalesce today. Each session builds on the last, and the demonstrations use Coalesce MCPs on live data across Coalesce Transform, Catalog, and Quality.
What the series covers
Part 1 | On- Demand
Getting Started With MCPs: Impact Analysis and Root Cause Debugging With Prompts
Watch a single MCP prompt trace column-level lineage, read Catalog ownership, and check Quality status to answer ‘what breaks if I change this field?’ Then follow a live root cause investigation from anomaly alert to the commit that caused it, and see where the guardrails sit before the team scales up.
Part 2 | On-Demand
Repeatable AI Workflows: Build Your First Data Engineering Skill With MCPs
Package the report you rewrite every Monday into a skill your team builds once. Go from a blank SKILL.md file to a weekly data quality report that pulls live data from Coalesce Quality, posts to Slack, and holds up to assertions, then hand it off to a triage skill that drafts the tickets.
Part 3 | On-Demand
Putting AI to Work in Production: Playbooks and Built-In Agents for Governed Data Workflows
A skill handles one task; a nine-phase governance rollout or 38 simultaneous alerts is a workflow. See two production playbooks that encode who decides, who executes, and what happens when the happy path breaks, plus where Coalesce Copilot and Scout fit as built-in agents instead.
Who should attend
Any data team putting AI to work on real data engineering, whether or not you use Coalesce. Data engineers, analytics engineers, and data and governance leads will get the most from it. Examples run in an MCP-compatible client like Claude Code, and every pattern shown works with any compatible client.
Instructor