SAP Business AI in 2026: From Applied AI to Agentic AI with Joule Studio

Jeet Poptani, Chief Transformation Officer at AumentoAI, revisits applied AI for SAP customers now that “applied AI” has become agentic AI running in production.

When this piece first ran, “Applied AI” for SAP customers meant machine learning models bolted onto SAP data for demand forecasting and predictive maintenance — useful, but external to the ERP itself. That framing is now out of date. SAP's own Business AI Platform, consolidated and re-architected through 2026, has moved AI from an adjacent capability to a layer of the platform itself, and the practical questions for an SAP customer have changed completely.

What the platform actually looks like now

SAP's Business AI Platform is best understood as three layers, and knowing which layer a capability sits in tells you who owns it, how it's priced, and what governance it needs.

The Context Layer is the foundation: the Generative AI Hub, SAP's own proprietary models (SAP-ABAP-1 and SAP-RPT-1), the SAP Knowledge Graph, and Business Data Cloud. Every other layer draws on this one, and it's priced on tokens — a real conversation-length interaction can run 5,000 to 20,000 tokens, which is a budgeting line most IT finance teams haven't had to model before.

The Build Layer is where agents get built: Joule Studio 2.0, generally available since June 2026, ships a visual Agent Builder, more than 2,500 pre-built SAP skills, CLI tooling for DevOps, and support for the open Model Context Protocol so agents can reach outside SAP's own catalogue. There are three ways to build here — assembling existing SAP skills (lowest risk), integrating external APIs (moderate effort), and grounding a custom model on your own data (highest complexity, highest differentiation) — and most enterprises should start firmly in the first category before earning their way to the third.

The Governance Layer is the part most 2023-era “applied AI” conversations never had to think about: the SAP AI Agent Hub, built on SAP LeanIX and launching in Q3 2026, gives a tenant-wide inventory, verification and policy management view across every agent running in the estate — which is exactly the discipline the EU AI Act now requires for high-risk use cases.

The strategy to achieve results

The vendor pitch is fifty-agent catalogues and thousands of pre-built skills. The sequencing that actually delivers a return looks narrower:

1. Pick two or three Skill Assembly use cases against your highest-cost manual processes first — not the most impressive demo, the highest-volume decision. Skill Assembly is lowest-risk because you're consuming SAP-built and SAP-supported logic, not grounding a custom model on data you haven't yet cleaned.

2. Model the token economics before you commit, not after the pilot. A use case that looks free in a demo has a real, variable cost per interaction once it's running at production volume — build that into the business case alongside the productivity gain.

3. Stand up the governance layer in parallel with the first pilot, not after the fifth. The AI Agent Hub exists specifically because “we'll govern it once we've proven value” is how enterprises end up with an ungoverned agent estate and an EU AI Act compliance gap at the same time.

The technical know-how this actually requires

This is no longer a data science exercise sitting next to SAP — the capability now needs to sit inside the SAP delivery team:

• BTP and Generative AI Hub configuration — understanding how token consumption, model selection and data grounding actually work, because that decision now sits inside your SAP landscape, not in a separate ML platform.

• Joule Studio's Agent Builder and skill assembly patterns — the practical skill of composing existing SAP skills into a working agent, which is a materially different competency from traditional ABAP development.

• Model Context Protocol integration — for the moment a use case needs to reach beyond SAP's own skill catalogue into other enterprise systems.

• AI Agent Hub and LeanIX-based governance — inventorying agents, defining approval workflows, and producing the audit trail an EU AI Act risk classification requires.

• A genuinely clean core underneath all of it — every one of these capabilities assumes standard data models and standard processes. The gap between the demo and your production reality is still, overwhelmingly, a data readiness gap, not an AI capability gap.

One practical note for anyone building now: Joule Studio 2.0's free design-time access ends at the close of 2026. Every agent built between now and then creates a technical dependency on separately licensed runtime capacity from 2027 — worth building into this year's roadmap while the negotiating position is still yours.

AumentoAI advises CIO and COO offices on sequencing SAP Business AI Platform adoption — from first Joule agent to a governed, audit-ready agent estate. Book a Value Advisory Session to map your Business AI Platform roadmap.