Agentic data organisation with Claude

In one line: Claude coordinates agent roles (analytics, platform, governance) — not a human org chart — so data delivery has clear ownership and quality gates.

Context

Classic data teams collapse analytics, platform, and governance onto the same people. I wanted an operating model that could run faster without losing traceability.

Why it was worth building

Without explicit roles, agent-assisted data work becomes a single chat thread with no accountability. Naming agent roles makes handoffs and gates testable.

What I built

Coordination layer

• Claude — prioritisation, cross-team summary, risk assessment (coordinator, not “Head of” people).

Agent roles (specialist workers, not employees)

• Analytics lead agents — KPI ownership, insight generation, dashboard agents.
• Platform lead agents — pipeline engineering, infrastructure reliability.
• Governance lead agents — quality, policy, compliance checks.

Daily loop — signal collection → leadership summary → task distribution per lead → quality gates → memory feedback. High-risk paths escalate to governance agents before release.

This project page is the canonical case (the former blog post redirects here).

How it was adopted

Used as the reference model for agentic data delivery on this site and related platform work — always with human goals and approval outside the agents.

Effect

Clearer ownership map for agent-assisted analytics and platform tasks; faster decision cycles than undifferentiated “ask the model” workflows.

Where AI fits now

This is the agents on data complement to governed lakehouses: MCP-scoped tools, evals, and enterprise platform boundaries. Coordination moved toward GitHub + Cursor Automations for this repo, but role-scoped agents and gates remain the durable idea.