You just became Head of Data Products. Here is a 90-day plan that runs on agents.

Look at the open roles at large enterprises right now and a pattern jumps out. Insurers are hiring a Head of Data Products inside a global data office. Pharma companies are hiring commercial data product owners. Media groups have data product managers in four different orgs. Banks are building whole job families for data product managers across every line of business.

The mandate is the same everywhere: turn scattered tables into products people trust and reuse. The trap is the same too. The new leader inherits a backlog of broken pipelines, conflicting definitions, a catalog nobody updates and an access queue, and spends the first two quarters firefighting instead of shipping.

Here is a 90-day plan that avoids the trap.

Days 1 to 30 — see the whole estate without asking for it

Connect Data Workers to the warehouse, orchestrator, catalog and BI layer over MCP. No migration. Agents start at L1, observing only.

By the end of the first month, the agents have mapped what most new leaders spend a quarter discovering:

• •Every pipeline, table and dashboard, with owners where they exist and gaps where they do not.
• •Lineage from source systems to the reports executives read.
• •Which definitions conflict: three versions of "active customer", two of "revenue".
• •Where the time goes: incidents, reruns, access requests and schema changes by domain.
• •What the agents would have fixed on their own, with receipts.

That last list is your business case.

Days 31 to 60 — automate the back office, domain by domain

Pick the two domains with the most toil. Open autonomy where it is safe:

• •L4, fully autonomous: catalog and documentation upkeep, sandbox cleanup, cost fixes, routine access grants.
• •L3, act and reversible: reruns and backfills, schema drift fixes in staging, quality rules on new tables, PII tags.
• •L2, propose: anything that changes an executive or regulated number. Held for a named approver.

Turn on standing routines in Autopilot: freshness, quality, cost and catalog upkeep. Set one org standard, such as "every certified data product has an owner, an SLA and a governed definition", and let the agents keep it true.

Days 61 to 90 — ship the first real data products

With the back office running, the team has its time back. Use it on two or three data products the business has asked for and never got:

• •Publish governed definitions in Data Context Wizard so copilots, analysts and agents all answer from the same meaning.
• •Give each product a living page in Spellbook Data Catalog, with owner, SLA, lineage, quality and change history kept current by agents.
• •Track autonomy on the Autopilot dashboard: what share of each team's work now runs on its own, and at what token cost.

What you walk into your 90-day review with

• •A map of the whole estate and the definitions that matter.
• •Two domains where most back-office work runs without people.
• •Two or three shipped data products with owners, SLAs and governed meaning.
• •A cost and autonomy dashboard you can show the CDO.

That is a very different first quarter from "we documented the problems".

Read next: the CDO data product office playbook.