Enterprise architect Harish Parmar says useful enterprise AI needs more than clean data: it needs context about how a business works, plus governance built in from the start. His blueprint gives organisations a concrete way to think about preparing AI agents to work across their systems, rather than simply adding a chatbot and hoping it understands the filing cabinet.
SAP Watch analysis
What happened
In an article published by SAP News, Parmar describes five layers of context for enterprise AI: data; semantic and knowledge; architecture and governance; processes; and company-specific policies and rules. Together, they are meant to help AI connect information across systems and understand what it is allowed to do.
The article gives the example of a customer whose records use different identifiers in SAP, Salesforce and a point-of-sale system. A unified customer record could help AI see those interactions as belonging to the same person. For governance, Parmar says organisations should document which agents operate, which systems they can access, which models they use and how decisions can be reviewed. Read SAP’s article.
Parmar also recommends treating custom agents as independent products, making them configurable and able to communicate through standard protocols such as MCP, and designing them as headless services where practical. He says human accountability remains non-negotiable.
Why it matters
An agent working across business systems needs more than access to data. It needs to know what that data means, which processes it belongs to, and where permissions and approval limits apply. Otherwise, “autonomous” can become a very polished way to make the wrong change in several places at once.
The advice also makes AI readiness an architecture and governance problem, not just a model-selection exercise. That is useful for organisations that cannot replace their existing systems overnight and need agents to work across SAP and non-SAP tools.
Our read
The strongest part of this blueprint is its insistence that context and governance belong in the design, not in a tidy-up after the agent has already been given the keys. The five layers are a practical checklist, though the article is guidance from an SAP architect, not evidence that adopting them guarantees successful AI projects. Start by mapping what an agent would need to know and what it should never be allowed to do.
What to watch
- Whether organisations turn these context layers into specific architecture and access controls.
- How agents will work across systems during migrations, including SAP ECC to S/4HANA.
- Whether organisations can show that governance and review work in deployed systems, not just on diagrams.
Discussion spark: Should companies be required to map an AI agent’s data, permissions and business context before deployment, or is that too much overhead for early experimentation?
Sources and evidence
- Why Enterprise Architecture Is Becoming Key to Successful AI (20 November 2026, 11:15 UTC)
Independent WittyWires tracker for public updates about SAP. Not affiliated with or endorsed by SAP; this is not an official account.