events

Governed Data for Agents Tanzu's AI-Ready Data Foundations

The hard part of enterprise AI was never the model. It's letting an agent use enterprise data without giving it access to enterprise data. Tanzu's answer is governed data products published to a marketplace. GA is Fall 2026.

Every enterprise AI conversation hits the same wall eventually. The model is the easy part. An agent is only useful if it can reach your data, and your data is the thing you can't let it reach unsupervised.

Broadcom's answer, announced on the opening day of Explore 2026, is a set of AI-ready data foundations in the Tanzu Platform. Purnima Padmanabhan, VP and GM of the Tanzu Division, described it as pairing governed agents with governed data inside the customer's own boundary.

This was one of seven announcements dated 31 August. It got less attention than the infrastructure pieces, which is the usual pattern and usually the wrong way round.

One thing up front. These capabilities are slated for general availability in Fall 2026. Plan for it, don't deploy against it.

The data product model

Data owners build and manage pipelines across structured and unstructured data, producing data products shaped for agent consumption. Those products publish to the Tanzu Platform marketplace as governed services that agents and developers can find and use, without data leaving the enterprise.

Three things in that are worth pulling apart.

The unit of governance is the data product rather than the query. That's a real choice. Governing at query time means evaluating every request against policy, which is expensive and brittle. Governing at product level means the shape and scope of what an agent can see gets decided once, by somebody accountable, and then reused.

Structured and unstructured are handled together. Most data governance tooling does structured data well and unstructured data badly, which is a problem when the documents are where the useful context lives.

And there's a marketplace. Discoverability is underrated here. A governed data product nobody can find gets bypassed, and the bypass is what ends up in the incident report.

Optimised for low-cost consumption

Broadcom describes the data products as optimised for low-cost agent consumption. That's a token economics point rather than a governance one, and it connects to the tokenomics thread running through the whole launch.

In practice: an agent reading and reasoning over raw source data burns far more tokens than one handed a pre-shaped product. That difference compounds with every invocation. So data product design becomes a cost lever as well as a governance one, which is an unusual thing to be able to say about a data engineering decision.

Where it sits

This lands in the Tanzu Platform, positioned as the agent platform for Private AI Cloud. The gateway principle from the agentic security announcement applies here too. Agents don't get direct access to data. They get access to a data product through a gateway that binds the connection.

So the chain runs like this. AI Factory provisions the infrastructure and runs the models. Tanzu produces and governs the data products. AgentMinder governs the agents and what they can invoke. vDefend and Avi cover the network. Four separate announcements, one control chain.

What to do now

• Work out who owns data products. This role probably doesn't exist in your organisation. It isn't the DBA or the data engineer or the platform team. It's whoever is accountable for what an agent is allowed to know.

• Inventory the unstructured data. If the useful context lives in documents, wikis and ticket history, that's the corpus that matters, and it's usually the one nobody has catalogued.

• Treat data product design as a cost decision as well as a governance one. Frame it that way early, because it changes who needs to be in the room.

• Plan for Fall 2026 without committing to it. Announced availability windows move. The organisational work ownership, inventory, policy doesn't depend on the product shipping.

The organisational work is harder than the technical work here, and you can start on it today.