What Zhamak Dehghani's Data Mesh Taught Our Catalog Agent
Four principles that changed how we think about catalog federation — and why 'who owns this data?' is a better first question than 'which catalog holds this table?'
By The Data Workers Team
In 2019, Zhamak Dehghani published an essay on Martin Fowler's site that changed how a lot of practitioners think about data platforms. The title was 'How to Move Beyond a Monolithic Data Lake to a Distributed Data Mesh.' The argument was deceptively simple: centralizing data ownership does not scale, because the bottleneck is not the technology — it is the organizational structure that puts a single team in charge of data from every domain in the company.
The First Question Is Wrong
Most catalog tools — and most catalog agents — start with the same question: 'which catalog holds this table?' It is a reasonable question if you treat catalogs as databases of metadata. But Dehghani's framework suggests a better first question: 'which domain owns this data, and have they committed to its quality?'
As Dehghani wrote: 'Instead of flowing the data from domains into a centrally owned data lake or platform, domains need to host and serve their domain datasets in an easily consumable way.' That shift — from ingestion to serving, from central to domain — changes what a catalog is for.
What Is Actually Worth Learning
- •Domain-oriented ownership: every dataset has a domain team accountable for it. A dataset that crosses domain lines without declared ownership is a governance gap, not just a metadata gap.
- •Data as a product: a dataset is not production-ready just because it exists in a catalog. Dehghani named six qualities a data product must have: discoverable, addressable, trustworthy, self-describing, interoperable, and secure.
- •Federated computational governance: the governance model balances what must be global (interoperability standards, classification tiers) with what should stay local (domain SLOs, ownership).
- •Shifting accountability upstream: 'The accountability of data quality shifts upstream as close to the source of the data as possible.'
How a Method Becomes a Skill
The domain-oriented-data-ownership skill encodes Dehghani's method as a different retrieval order. The agent first resolves the domain boundary, then fetches the dataset profile, then runs the six-quality product fitness test, then pulls the lineage graph to check for undeclared cross-domain dependencies. The report is organized by domain owner, not by catalog. Datasets that fail the fitness test are flagged as 'exists but is not yet a data product.'
One of More Than 400
The domain-oriented-data-ownership skill is one of more than 400 method-named skills across 19 agents in the Data Workers swarm.
A note on this post: This is independent commentary and homage. It distills publicly available writing and talks by Zhamak Dehghani to illustrate a working method, and every quote is drawn from and verified against the primary sources linked above. The skill it describes is named for the method, not the person, and contains no marketing claims attributed to them. Data Workers is not affiliated with, sponsored by, or endorsed by Zhamak Dehghani. If you are Zhamak Dehghani and would like anything adjusted or removed, email hello@dataworkers.io and we will respond promptly.
Related Posts
What Ralph Kimball's Dimensional Modeling Taught Our Pipelines Agent
Ralph Kimball's four-step dimensional design process is one of the most durable ideas in data engineering — here is what it taught our pipelines agent.
What Jay Kreps's Log-Centric Architecture Taught Our Streaming Agent
Jay Kreps's core insight is deceptively simple: an append-only, totally-ordered log is not just a message bus — it is the single source of truth that eliminates N² integration pipelines and makes reprocessing routine. We studied his published writing and built a reusable streaming skill around the method.
What W. Edwards Deming's Plan-Do-Study-Act Taught Our Data Quality Agent
W. Edwards Deming spent a career arguing that quality comes from improving the process, not inspecting for defects. His Plan-Do-Study-Act cycle is the most rigorous improvement loop in the field. Here is how we encoded it into our data quality agent.