Everyone shipped data products. The decision rights never left the building.

Data mesh asked for two things: treat data as a product, and move the decisions to the people who own the domain. Organizations did the first enthusiastically. The second is still pending — which is why data products multiplied while accountability stayed exactly where it was.

What happened: the vocabulary of data mesh won; its operating model lagged. Zhamak Dehghani's original formulation pairs "data as a product" with domain-oriented ownership, inverting responsibility so quality accountability moves upstream to the domains. Thoughtworks' own 2026 retrospective reports the recurring anti-pattern: rebranded IT teams called "domains" lacking a clear mandate, aligned incentives, or the authority to make decisions.

Why it matters: a data product is a delivery wrapper — a dataset with an owner, an SLA, and decision rights attached. Strip the decision rights and only the wrapper remains: a landing page and a roadmap nobody can enforce. The catalog looks like progress because it is countable. Decision rights are not countable and require someone senior to give authority away. The measurable half ships; the rest waits.

What it reveals about the field: we reliably adopt the artifact and skip the authority. Thoughtworks has said since 2020 that the problem is organizational, not technological; in 2026 the greatest obstacles are still behaviors, not architectures. The platform did not fail to create ownership. It revealed there was none — and federated governance without federated decision rights is a steering committee with a new logo.

What to watch: watch for the first real decision a domain may make alone — deprecate a field, reject a consumer's request — without escalating. If it still routes to a central team, the mesh is topology, not ownership. The tell is a domain that can publish a data product but cannot say no to one.