1. The ERP Brain Has a New Job. It Works for Your Agents Now.

SAP Business AI Platform

Almost every company is using AI. Almost none are getting the business value they expected. The Stanford AI survey Christian Klein referenced at the open says what many CIOs already know privately: the problem isn't the model. The problem is everything below the waterline: 7 million data fields, thousands of proprietary processes, authorization rules, governance requirements. None of which any model was trained on, and none of which any model naturally respects.

SAP's answer is the SAP Business AI Platform, launched at Sapphire 2026. Its core is a rich context layer that injects ERP domain knowledge directly into every agent interaction. Knowledge graphs navigate thousands of business processes and select the right data. SAP Domain Models give agents native understanding of business logic. Business Data Cloud provides a single semantic layer across SAP and non-SAP systems. The governance layer checks identity and authorization before any output surfaces.

JPMorgan Chase joined the keynote to make this concrete. The bank processes 12 trillion dollars in payments across 120 currencies every day and is already deploying agentic solutions that catch errors before they post to the ledger. Jeremy Barnum, CFO, JPMorgan Chase:

“AI is only as good as the data and processes underneath it. It can't reach its full potential in a fragmented legacy environment.”

The framing is deliberate. SAP is not adding AI to ERP. It is bringing 50 years of ERP domain knowledge into AI. That is a different claim, and a structural one.

2. 224 Agents, 51 Assistants. The Pilot Era is Over.

The Autonomous Suite

2. The Autonomous Suite

Enterprise AI has lived in pilot purgatory for two years. Impressive demos, limited deployment, unclear ROI. The structural problem is that most agentic platforms are horizontal: they give you a foundation and then require you to build your own business process agents from scratch. SAP is taking the opposite position.

The Autonomous Suite launches with 224 pre-built agents and 51 assistants across five domains: Autonomous Finance, Autonomous Spend, Autonomous Supply Chain, Autonomous HCM, and Autonomous CX. These are outcome-driven, SOX audit-compatible, ISO-certified agents mapped to specific business roles with defined KPIs. Each is extensible: you can add tools, workflow steps, or custom agents for what does not fit out of the box.

Early customers are already in production. Bayer is using the Cash Collections Assistant to accelerate cash flow. Ericsson deployed intelligent role recommendations across 85,000 employees. Petco is using the Merchandising Assistant to surface assortment gaps and optimize product data across all channels. The number of agents and assistants will grow monthly.

3. Company Memory: SAP Decides to Solve the Tribal Knowledge Problem

Company Memory

SAP introduced Company Memory as a knowledge management layer for the Autonomous Suite. Built on the Signavio foundation, it ingests process models, policy documents, Teams and Slack channels, and email approval chains. From that unstructured material, it extracts “process atoms”: small structured units of business knowledge that tell agents what to look for, what to do, and what not to do. Every override, every pattern, every correction becomes instruction for the next run.

You can directly update any process atom manually, and every agent in your landscape adapts instantly. The system is not learning autonomously without oversight. It is learning with you. The Financial Closing demo illustrated this: 12 legal entities, four ERP systems, six currencies, hundreds of reconciliations. Q2's close was measurably faster than Q1. Each close teaches the next one.

4. Joule Gets a Ground-Up Rebuild. The Internal Name Was V10.

Joule Work

The current Joule is RAG-based, limited in context and governance. Joule Work introduces a full agentic harness with computer and file access, open standards support including MCP, and a more complete knowledge base. The new Joule Spaces interface creates contextual workspaces that generate dynamically based on the task at hand. SAP's framing is “App-less”: an experience layer where users interact with their underlying applications without having to be in them.

The Financial Closing demo gave Joule Spaces its clearest illustration. The Closing Assistant surfaced 18 items requiring human review inside a single workspace: accrual proposals with confidence levels, intercompany variances already resolved, one-click reconciliation across currencies and entities. The human stayed in the loop without ever opening a separate application. That is what “appless” looks like in practice.

5. Build Agents for Outcomes, Not Architecture: Joule Studio 2.0

Joule Studio 2.0

The new Studio is intent-based: you describe the business challenge, not the technical specification. Philip Herzig demonstrated this with a pharma distribution scenario: margin leakage of $24 million identified via SAP Signavio. Joule Studio took that diagnosis, generated a full product requirements document, identified two existing agents already covering part of the workflow, and built the full orchestration without writing a single line of code. KPMG is targeting $120 million in reduced contract leakage using this stack.

Joule Studio 2.0 is model-agnostic: bring any third-party LLM or SAP's own. Development is low-code, pro-code, or anything in between. The evaluation framework is data-driven, so you validate agent behavior against actual business outcomes. It is available free through end of year.

6. Industry AI: The Superpower SAP Has Been Sitting On

Industry AI

6. Industry AI

Sebastian Steinhauser opened with a raised hand. Who is using SAP 100% in standard? Almost none. Who is using SAP with industry-specific solutions? Most of the room. That gap has historically been filled by customizations and the expertise of people who have been in the industry for 20 years.

Industry AI is SAP's move to build that depth in. Seven autonomous industry solutions are launching now, spanning adaptive production, project delivery, regulated manufacturing, revenue growth management, and unified commerce. Take regulated manufacturing: in pharma, every batch must be traced and audit-ready. The Industry AI solution delivers multi-agent processes that detect disruptions early and simulate compliant recovery options in real time. Takeda reports up to 25% reduction in revenue loss from stockouts.

H&M offered the fullest picture. 4,100 stores across 81 markets, with a Store Intelligence Agent coordinating end-to-end across the value chain in real time. Ellen Swanstrom, Chief Digital and Information Officer, H&M:

“To unlock the full potential of new technologies like AI, the right digital foundation is needed to ensure that we can standardize and harmonize key business processes. And SAP sits at the core of our processes.“

7. Below the Waterline: Dremio, Prior Labs, and SAP-RPT-1.5

Business Data Cloud

SAP intends to acquire Dremio, a high-performance data lakehouse with federation across any cloud and legacy database. Combined with Business Data Cloud, this enables agent reasoning across any cloud environment without data movement. SAP also intends to acquire Prior Labs, whose TabPFN model family handles non-SAP structured data for prediction use cases including financial trading. And SAP is launching SAP-RPT-1.5, the next generation of its own tabular foundation model, with a chat API for what-if analysis, predictive simulation, and built-in explainability, without any model training required.

Together, SAP-RPT-1.5 and TabPFN address the part of enterprise AI that large language models were never designed for: accurate, explainable, prediction-grade outcomes on tabular business data. The LLM is the tip of the iceberg. The data layer being assembled here is everything underneath it.

8. One Command Center for Every Agent You Didn't Build: AI Agent Hub

AI Agent Hub

8. AI Agent Hub

 

By now, most large enterprises have more AI agents running than anyone has officially counted. Some are great. Some are not. Almost no one has a consistent picture of which agents are operating, what they are allowed to do, and whether they comply with governance requirements.

SAP AI Agent Hub, built on LeanIX (already used by more than 25% of the Fortune 500), is the central command center for all of it. Teams can discover, manage, and govern SAP agents, non-SAP agents, and MCP servers across the landscape. Telemetry routes to Cloud ALM for health monitoring. Signavio's agent mining surfaces optimization opportunities against KPIs. All agent activity maps to SuccessFactors so the workforce impact is visible end-to-end.

Three things worth noting: governing non-SAP agents is free. A2A interoperability with third-party agentic platforms is also free. And the full governance lifecycle is included in SAP Business AI Platform at no extra charge. The governance infrastructure is being given away. The value proposition is the agents running on top of it.

9. Rise and Grow with SAP: Change Management Written Into the Deal

Rise and Grow with SAP

The keynote offered a clear framing: “Technology alone and plugging AI agents into your existing system landscape will drive zero value.” That framing is deliberate. Rise and Grow with SAP is built to answer it directly.

The new Rise and Grow with SAP offering responds directly. SAP commits to sending architects, AI developers, and industry consultants on-site to drive change management and AI adoption, regardless of whether the customer starts on S/4HANA or ECC. The contractual commitment: activate three dual assistants of your choice within year one. For migration, Joule for developers and Joule for consultants are being extended with AI assistants targeting a 50% reduction in total migration effort.

The tooling behind that commitment is now visible. Seven specialized migration assistants, covering system analysis, data management, custom code, configuration, test management, rollout, and project management, are orchestrated inside Cloud ALM. Joule for Consultants is the single entry point across all of them. System analysis that previously required weeks of preparation can now run over a weekend.

Technology. Change management. Migration. These are the three failure modes of large-scale ERP transformation, and SAP is now naming all three in the same commercial offer. The argument is not that SAP's AI is the best AI. It is that no matter where you start, SAP will make you ready to use AI tomorrow.