The introduction of the SAP Business AI Platform and Joule Studio 2.0 is fundamentally rewriting the rules of system architecture. We are moving away from classical, rigid programming towards a declarative approach. This means we no longer tell the system how to perform a task step by step, but rather what the desired business outcome is.
Everything else — finding the right data, selecting the appropriate tools, and applying logical reasoning — is handled by the AI agent entirely on its own.
If the AI has that much autonomy, why do we still need an architect? The answer is simple: because AI systems are equipped with compliance knowledge, legal boundaries, and corporate strategy only through human expertise. This is where the Intent Architect (goal and prompt designer) comes in. They are no longer the micro-manager who prescribes every process step; they are the lawmaker who defines the playing field.
The Direct Comparison: Yesterday vs. Today vs. Tomorrow
| Dimension | The Classic Solution Architect (Yesterday/Today) | The New Intent Architect (Tomorrow) |
|---|---|---|
| Primary focus | Infrastructure & data flow: How do we connect system X to SAP S/4HANA? What interface is needed? | Semantics & guardrails: How does Joule understand the business intent? What behavioural boundaries do the agents have? |
| Process execution | Imperative: Rigid "if–then" logic, iDocs, and hard-wired process chains. Every click is prescribed by a human. | Autonomous & declarative: The agent selects its own solution path and tools (APIs) dynamically based on context. |
| Human role | Building bridges (interfaces) between systems. | Building the box (constraints & guardrails) within which the agent is permitted to operate. |
What Does an Intent Architect Actually Need to Know? (Hard Skills)
A "prompt engineer" who only writes creative text for ChatGPT will fail in an ERP environment. The Intent Architect needs deep SAP process knowledge combined with AI governance expertise:
1. Deterministic Prompt Engineering & Tool Provisioning
They must instruct LLMs within SAP AI Core to deliver reproducible, deterministic results. Their primary task in Joule Studio 2.0 is tool provisioning: they define precisely which "toolbox" (e.g. which OData APIs) the agent has available for a given intent. The agent then selects the right tool autonomously — but it can only see what the architect has authorised.
2. SAP Business Data Cloud (BDC) & Semantic Data Models
For the agent to draw logical conclusions, it needs context. The architect must understand and maintain the SAP Knowledge Graph, ensuring the AI knows how business objects (e.g. SalesOrder, Product, Supplier) relate to one another.
3. Integration Landscapes & API Security
Because enterprise landscapes remain heterogeneous, the Intent Architect governs how agents interact with third-party systems (e.g. Salesforce or non-SAP databases) through the SAP Integration Suite, establishing security and access patterns at the API level.
The Practical Example: Autonomy Within Hard Guardrails
Scenario: Senior management issues the directive: "Due to supply-chain disruptions in Asia, we need to prioritise all open semiconductor orders that can be rerouted via European hubs."
What the AI agent does autonomously:
The user enters the command. The agent starts its reasoning and decides dynamically:
- I call the material master API to identify "semiconductors."
- I check open purchase orders via the S/4HANA API.
- I search the Knowledge Graph for alternative routes through Europe.
- I calculate the new delivery dates.
None of these individual steps were pre-programmed by the architect. The agent selects its own skills and tools entirely autonomously.
What the Intent Architect has secured in the background:
To ensure this process runs in an orderly and legally sound manner, the Intent Architect has defined the sandbox inside Joule Studio 2.0:
- The tool boundary: They have specified: The agent may read freight-rate APIs and modify delivery dates for this intent. The master-data API for prices or bank details is blocked for this agent. (Prevents unauthorised financial transactions.)
- The legal emergency brake (four-eyes principle): A hard compliance rule is stored in the system: The agent may autonomously book route changes up to an additional cost of €5,000. Anything above that stops the autonomy and transfers the case as an exception to a human supervisor.
- Legally valid declaration of intent: Since an agent has no legal personality, the architect ensures that every autonomous booking in the SAP system is logged with a unique timestamp and the ID of the underlying approval. This makes the action legally valid as a company-authorised declaration of intent — audit-proof and traceable.
Conclusion for Companies
The Intent Architect is the chief translator of business logic. They trust the autonomy of the AI agents but control them through uncompromising, system-embedded guardrails, ensuring that strategic management decisions are executed flawlessly, in real time, and with full legal certainty.
For existing SAP Solution Architects, this role is the logical evolution: their process knowledge remains the foundation — steering through semantic intents is the new language.