Brand1.1
SAP Business AI is the umbrella, not a product.
An AI brand spanning 815 embedded use cases, with 240+ AI features and agents shipping in Q2 2026 alone. Not one thing — a stack.
Joule is the headline. AI Foundation is the engine room. But Business Technology Platform is the foundation slab — pull it out and the rest collapses. This is the reference map most blogs hand-wave through.
An AI brand spanning 815 embedded use cases, with 240+ AI features and agents shipping in Q2 2026 alone. Not one thing — a stack.
SAP AI Foundation = SAP AI Core + Generative AI Hub — the enterprise AI platform managing the full lifecycle of ML and GenAI models. Sitting alongside on BTP: Joule Studio (authoring) · SAP AI Launchpad (operations) · SAP HANA Cloud Vector Engine (data) · Document Grounding (inside AI Core) · SAP Document AI · Knowledge Graph.
Underneath: Skills (deterministic, rule-based) and Agents (autonomous, reasoning-driven). Authored in Joule Studio, consumed in Joule.
You cannot run Joule without a BTP global account, subaccount, SAP Cloud Identity Services (IAS + IPS), Cloud Foundry runtime, the Destination service, and SAP Build Work Zone. None of this is optional.
Base AI (free, embedded). Premium AI (PUPM with AI Units). AI Foundation (BTPEA credits, consumption-based). Almost every conversation lives in tier three.
What each piece is, how BTP knits them together, how to translate a real business scenario into a concrete stack — with diagrams, references, and a quick-start plan at the close.
Walk into any SAP customer in 2026 and ask "what's your AI strategy?" — you'll hear about Joule, AI agents, Generative AI Hub. What you rarely hear is the unglamorous truth that holds it together: none of it works without BTP. SAP says it themselves — AI Foundation is "an operating system to build, run, and integrate AI solutions at scale on SAP Business Technology Platform" (SAP Business AI 2026 deck, p.31). This brief is a practical companion to SAP's AI Golden Path — the canonical SAP track for AI development. We focus on the BTP-substrate foundation it assumes.
"How do I turn a business scenario into a concrete stack?" — the killer question every SAP architect actually wants answered. Five patterns. Pick the closest match.
Embedded Joule answers using built-in Skills. The entry door — if you can't get this working, none of the others will. Zero custom code. The starting point for every Business AI conversation.
// Assumes BTP Phase 1 in place. If greenfield, add 2–4 weeks for the foundation (see §2 Phase 1 callout).
A reasoning agent that ingests a PO, compares to expected, decides next action. Multi-step. Open-ended path. The signature use case for Joule Studio. Authored once, runs across procurement.
// * Assumes BTP Phase 1 in place. Add 2–4 weeks if greenfield (incl. Cloud Foundry runtime and Destinations to S/4HANA).
Document-grounded Joule for compliance Q&A. The fastest custom win — minimal code, real value, immediate audit utility. Document Grounding orchestrates ingest and chunking; embeddings are stored in SAP HANA Cloud Vector Engine (not a separate vector database). Everyone in the org needing policy lookup becomes self-service.
// Assumes BTP Phase 1 in place. Document Grounding requires AI Core subaccount; allow 1–2 weeks if not provisioned.
RAG over operational data using HANA Vector. Native SQL COSINE_SIMILARITY and L2DISTANCE with relational filters in a single query. No separate vector DB. Embeddings sit alongside business data — semantic + structured query, one engine.
// * Assumes BTP Phase 1 + HANA Cloud provisioned. Greenfield BTP and HANA capacity sizing add 3–4 weeks (see §7 P.03).
Document AI for AP automation. Geographically distributed processing. EU-only access available for compliance use cases — important for regulated industries that must keep document data resident.
// Assumes BTP Phase 1 in place. EU-only mode requires region selection at AI Core subaccount creation — design decision before Phase 1 begins.
Every architect in 2026 is asking the same two questions: "How do we keep track of all the agents being built?" and "How do SAP agents collaborate with Google / Microsoft / AWS agents we already have?" SAP's answer pairs AI Agent Hub in SAP LeanIX (an Agent Radar with classification seals — Approved / Conditional / Requested / Denied) with the Agent2Agent (A2A) protocol (planned H1 2026) — a connector + ORD Aggregator that carries TRUST and ORD signals between SAP Business AI and Google Vertex/ADK, Microsoft Azure AI Foundry, AWS Bedrock. Pro-code partner agents integrate via SAP's named pattern: Bring Your Own Agent (BYOA). The agent exposes an A2A server endpoint, and an IAS App2App trust relationship establishes mutual authentication between Joule and the agent server. Pre-production trust evaluation feeds the classification. Approved agents get A2A clearance.
// * A2A protocol announced for H1 2026 (SAP Business AI deck pp.28–29). AI Agent Hub via LeanIX is available now. Wire your trust-evaluation framework today; A2A clearance is the activation moment.
Source · SAP Joule Studio overview / slide 03 / 2026
"Agents do not adhere to your security, compliance, and responsible AI needs."
SAP, framing the problem they say Joule Studio solves — and the gap they leave you to close.
"We'll figure BTP out later" turns into a six-month delay. Design it first.
Federating later is much harder than federating right.
Embeddings consume meaningful storage. Right-size at the start.
Defaulting to whatever LLM is cheapest creates unpredictable data-residency exposure.
Without governance, stale or out-of-band docs surface in answers.
Agent governance is harder. If the path is known, write a Skill. Save Agents for genuinely open problems.
"We'll add logging later" — until your auditor asks for the AI decision log and you have nothing.
Joule Agents take action. The audit trail and approval workflow matter as much as the prompt.
SAP's readiness checklist makes the Establish a Unified (Central) Joule Instance mission a mandatory pre-req before any LoB activation. Skipping it creates fragmented Joule footprints that are painful to consolidate later.
Premium AI is per user per month plus AI Units consumed by agents. Customers routinely underestimate AI Unit burn rate once agents are live. Plan capacity, set thresholds, monitor with AI Launchpad.
A fractional CISO and SAP architecture practice for mid-market SAP customers. Three things we do well — and a sixty-minute conversation that's actually useful.
Assess your current BTP, IAS, AI Core posture against the readiness required to deploy Joule and AI Foundation safely. Result: a prioritized roadmap.
Define the role model, model routing policy, document grounding governance, and agent approval workflow that sits underneath your rollout.
Apply our NGIQ-ATE methodology to evaluate Joule Agents and custom agents against six trust domains before they go to production.
This is a living architecture reference. When SAP ships material changes to the platform we cover here — at Sapphire, TechEd, or in a release-highlights post — we log the delta below with citations. The body of the brief is not rewritten unless a foundational assumption breaks; updates accrete here so readers can see what changed, when, and against which SAP source.
At Sapphire 2026 (May 12–15, Orlando) SAP announced a strategic reframe to "Autonomous Enterprise," a consolidation of BTP + Business Data Cloud + AI Foundation into a single SAP Business AI Platform anchored on the new SAP Knowledge Graph, the rollout of Joule Studio 2.0, an order-of-magnitude jump in agent inventory, a UX-layer reframe via Joule Spaces, and a series of partner formalizations on the data and identity layers. Twelve material shifts in total. We are not rewriting the brief — the Phase 1 BTP prerequisites, three-tier pricing, five scenarios, and trust boundary map all hold. We are annotating the affected sections, right-sizing one over-hyped interop story, and adding new architectural primitives (Knowledge Graph, Domain Models, Company Memory). Each entry is tagged by severity AND timing — when SAP announced it vs when customers can actually consume it. Severity legend: NEW primitive, SUPERSEDES a prior fact, UPGRADED from roadmap to GA, ANNOTATE existing frame, REFRAME strategic positioning, or RIGHT-SIZE a marketing-vs-reality gap worth calling out honestly.
Share this brief
Service icons adapted from the SAP BTP Solution Diagrams library (Apache License 2.0). SAP product names are trademarks of SAP SE.