SAP SAPPHIRE 2026: What Was Announced, and What's Changed

In May SAP declared the end of software that “records what people do” and the start of software that “does the work itself.” More than three months on, here’s what was announced then, what SAP has since confirmed in its quarterly results, and what independent analysts are saying about the gap between the vision and enterprise reality.

Analysis | ~18 min read | By J. Torre

In brief

At SAPPHIRE 2026 — Orlando (May 12–14) and, a week later, Madrid (May 19–21) — SAP CEO Christian Klein unveiled the “Autonomous Enterprise” vision, anchored by a new SAP Business AI Platform and the SAP Autonomous Suite. In May, SAP was talking about more than 200 AI agents and over 50 Joule Assistants across Finance, Spend, Supply Chain, HR and Customer Experience.
SAP Business AI Platform (BAIP) is not a product built from scratch. It is the planned consolidation of SAP Business Technology Platform (BTP), SAP Business Data Cloud (BDC) and SAP’s AI Foundation into one platform — SAP confirms existing BTP investments, integrations and competencies carry forward with no forced migration.
Execution is advancing, but still partial. On stage in May, the precise figure SAP gave was 224 agents and 51 assistants; on its Q2 earnings call (July 23), Klein raised the target to roughly 50 assistants by end of Q3 and 400-plus agents by year-end.
Anthropic’s Claude is the primary reasoning model behind Joule, alongside SAP’s own tabular model. The Prior Labs and Dremio acquisitions, pending in May, closed in July 2026 — Forrester still flags the Anthropic dependency as a genuine concentration-risk question for regulated industries.
Joule Studio 2.0 and the agent runtime remain free through December 31, 2026, but on the July 23 earnings call, Klein signaled the commercial direction: moving from per-user licensing toward pricing tied to the value agents deliver, with no 2027 price list yet.
The German-speaking user group DSAG has pushed back on one specific point: some AI services on BAIP remain coupled to RISE with SAP contracts, and DSAG is asking that on-premises S/4HANA customers get equal access.
Madrid added a layer that didn’t appear in Orlando: data sovereignty. EU AI Cloud, sovereign models from Mistral AI and Cohere, and a staged-autonomy, human-in-the-loop governance model, aimed at regulated sectors and European public administrations.

Why this matters now

If you run SAP, you have probably heard some version of “agentic AI” in every vendor call for the past year. SAPPHIRE 2026 is different in one respect: it is the first time SAP has put a name, an architecture, and a delivery timeline on the idea, rather than a roadmap slide. Constellation Research analyst Holger Mueller called it

“the first time on this side of the millennium that SAP has a vision for ERP,” in his analysis published on Constellation Research. His framing is worth sitting with: the question for SAP customers is no longer whether to upgrade to S/4HANA, but whether the autonomous enterprise value proposition is strong enough to justify staying inside SAP’s stack for the next decade.

This piece is organized in three parts, marked throughout the text, that separate three things that tend to blur together in vendor coverage: what SAP officially announced, what is shipping today versus what is still a promise, and what reputable independent analysts and SAP’s own user community are saying about the distance between the two.

PART 1 · WHAT SAP ANNOUNCED

What SAP announced on stage

In his keynote, Klein framed the problem SAP is trying to solve as a gap between AI investment and AI value. Citing a Stanford AI survey, he noted that AI adoption is now near-universal across companies, yet tangible business value remains elusive for most of them. His diagnosis: general-purpose large language models are excellent at tasks trained on public data, but enterprises need something different — AI that understands mission-critical business data, end-to-end processes, and operates inside security, compliance, and governance constraints.

“Eighty percent is just not good enough when you run the world’s most business-critical businesses,” Klein said, according to SAP’s own News Center coverage of the keynote. “They [LLMs] should not guess; they should deliver accurate, compliant, and secure outcomes.”

SAP’s answer, in Klein’s words, starts with what he called “the brain of every company: its ERP system.” The pitch is that 50-plus years of process and data know-how, encoded in SAP’s ERP systems, is the missing context that makes generic AI agents unreliable in business settings. Three announcements carried that argument on stage.

1. SAP Business AI Platform (BAIP)

SAP Business AI Platform is a unified architecture bringing together SAP Business Technology Platform, SAP Business Data Cloud, and SAP’s AI Foundation. SAP CTO Philipp Herzig described it as structured around three layers:

Context layer — built on SAP’s Knowledge Graph and new SAP Domain Models (trained on SAP’s own code and business logic), plus SAP Business Data Cloud, which unifies SAP and non-SAP data into what SAP calls a single semantic layer, avoiding what Herzig described on stage as “spaghetti data sprawl.”
Build layer — centered on Joule Studio 2.0, an intent-based environment where teams draft a requirements document and generate a contextualized agent or workflow from it. SAP is investing €100 million in partner ecosystem development for agent building, and Joule Studio 2.0 itself is free, with a no-charge offer on the agent runtime through year-end 2026.
Governance layer — anchored by the new SAP AI Agent Hub, built on SAP LeanIX, which SAP positions as a vendor-agnostic command center to discover, monitor and govern both SAP and third-party AI agents. It reaches general availability in Q3 2026 and is included in SAP Business AI Platform at no additional charge.

2. SAP Autonomous Suite

SAP Executive Board Member Muhammad Alam described the transformation of SAP’s SaaS application portfolio into the SAP Autonomous Suite as “the most significant evolution of SAP’s applications business in the company’s history.” The suite spans five domains, each built on out-of-the-box agents that roll up into role-based Joule Assistants:

Autonomous Finance — Financial Closing, Financial Planning, Billing, Governance, Tax and Compliance, Accounts Receivable, and Cash and Treasury assistants.
Autonomous Spend — procurement, sourcing, supplier management, invoicing and travel.
Autonomous Supply Chain Management — Product Design, Manufacturing, Asset and Service, Planning, Logistics and Business Network assistants.
Autonomous HCM — Core HR, Payroll, Time, Recruiting, Onboarding, Learning, Performance, Career and Skills assistants, targeting June 2026 general availability.
Autonomous CX — marketing, commerce, sales and service assistants coordinating multi-step workflows across SAP and third-party systems.

Each assistant carries a defined set of ROI KPIs tracked through SAP AI Agent Hub. SAP’s own figures put the footprint at over 200 agents and 50-plus assistants; Forrester’s post-event analysis cites the more precise figures SAP disclosed on stage: 224 agents and 51 assistants, developed under an ISO-certified process designed for SOX audit compatibility — a signal, as SAPinsider notes, that SAP is building this for CFOs and chief compliance officers as much as for CIOs.

3. Joule Work

The third pillar is a redesigned user experience. Joule Work combines the familiar Joule chat interface, Joule Studio 2.0, and a new agentic harness with computer and file access, and support for open interoperability standards — Model Context Protocol (MCP) and Agent2Agent (A2A). Alam described it as SAP’s attempt to make Joule the primary way people and systems interact with the entire SAP stack, rather than a chatbot layered on top of existing screens.

Industry AI and named customer proof points

SAP COO Sebastian Steinhaeuser introduced Industry AI, sector-specific solutions across 26 industries. On stage, SAP cited its life sciences customer Takeda achieving up to 10% productivity gains and up to 25% reduction in revenue loss from stock-outs through Autonomous Regulated Manufacturing. H&M Group CDIO Ellen Svanström demonstrated a Store Intelligence Agent and an AI-powered InStore Concierge built on RISE with SAP, SAP Business Data Cloud, SAP Commerce Cloud and SAP SuccessFactors. JPMorganChase CFO Jeremy Barnum appeared on stage confirming the bank is migrating its general ledger to SAP’s latest version, and Forrester notes Bayer, Novartis and Ericsson were cited as production references as well.

The most concrete third-party number came from a services partner, not SAP itself: on stage and in SAPinsider’s coverage, KPMG Global Head of Advisory Rob Fisher described deploying 20 agents on the new platform for a single client, targeting $120 million in reduced contract leakage. Fisher’s framing of the broader market shift is worth quoting directly: “What I’m hearing from clients is a clear shift; they’re moving from AI pilots to embedding integrated AI and agents into how work gets done. Where we see leaders really separating from the pack is in the execution and the organizational adaptability.” That is a services-partner incentive talking as much as an independent data point, but the specificity — a named client target, a dollar figure, an agent count — is more concrete than most of what got said on stage.

Company Memory: the quieter announcement

Easy to miss amid the platform and suite announcements was Company Memory, built on the SAP Signavio foundation. SAPinsider describes it as a knowledge-management and context-graph layer that continuously ingests policy documents, process models, and team communications — including chats and email approval chains — into structured “process atoms” that condition how agents behave. Every exception an employee handles updates the memory; every override becomes training signal for the next agent interaction.

The pitch is that this captures the tacit, undocumented knowledge that walks out the door when a long-tenured employee leaves. Forrester considers horizontal AI platforms from hyperscalers unable to replicate this without SAP’s decades of embedded ERP context, but flags an important caveat: there were zero public production references for Company Memory at the time of announcement. It is a compelling mechanism on paper, not yet proven in a named customer environment.

SAPPHIRE Madrid: the European sovereignty layer

A week after Orlando, SAP repeated much of the same Autonomous Enterprise narrative at SAPPHIRE Madrid (May 19–21, IFEMA), but with an emphasis absent from the US keynote: data sovereignty. For a European audience — particularly public-sector and regulated customers across Spain and the wider EU — SAP added three concrete pieces to the autonomous enterprise pitch.

EU AI Cloud — an AI infrastructure offering hosted and operated within the European Union, for customers who need AI data processing to stay inside European jurisdiction.
Sovereign models — agreements with Mistral AI and Cohere to offer model alternatives trained and operated outside the US-vendor axis, for workloads where model sovereignty, not just data sovereignty, is a requirement.
Orchestration with n8n — SAP confirmed in Madrid a strategic investment in n8n, the workflow-automation platform, as an integration layer between Joule agents and third-party systems.

The underlying message in Madrid was “staged autonomy”: a governance model where agents operate at different autonomy levels depending on task risk, with a mandatory human in the loop for higher-impact decisions. It is a direct, if not always explicit, answer to the question that comes up most with European public-sector and financial-services customers: who is accountable when an agent gets it wrong. For SAP leaders reading this from Europe, this layer — data and model sovereignty plus staged governance — is arguably more immediately relevant than much of what was announced in Orlando, precisely because it answers regulatory requirements that already exist today, not a future value promise.

From BTP to BAIP: what changes in practice

For anyone who has spent the past few years building on SAP Business Technology Platform, the most consequential announcement may be the least dramatic-sounding one: BTP is being folded into SAP Business AI Platform.

According to reporting gathered from SAP Community and partner-facing sources, the rename and consolidation takes effect by June 30, 2026, and SAP has stated that all existing BTP investments, integrations and competencies carry forward automatically — partners holding the BTP competency were transitioned to the BAIP competency with no action required on their part, and SAP has been explicit that this is not a forced re-platforming.

Whether this is “just a rename” depends on who you ask. SAP’s own framing, echoed across its partner communications, is that this reflects a genuine architectural shift: AI is no longer a layer added on top of BTP, but the organizing principle of the platform itself. SAPinsider’s practitioner-focused analysis describes it as “the architectural rationalization that was overdue” after two years of SAP’s AI portfolio expanding in multiple, loosely connected directions.

But the rename has not been received uncritically. The SAP user group DSAG has raised two specific, recurring concerns about how AI capabilities are being packaged inside BAIP:

Cost model — DSAG has criticized the cost of developing, quality-assuring and operating BTP/BAIP services under a consumption-based pricing model, which can make budgeting for AI initiatives harder to predict.
RISE coupling — DSAG’s central objection is that some AI services on the platform remain tied to RISE with SAP contracts. Because RISE is a specific commercial and technical model, this effectively restricts full AI access for S/4HANA customers who have not signed a RISE contract — including many on-premises customers. DSAG has stated it sees no technical reason for this coupling and wants AI functionality available to on-premises customers through the platform directly.

This is not a new fault line. Diginomica’s coverage of DSAG’s position ahead of previous SAPPHIRE events shows the user group making the same argument — that AI value should be decoupled from the RISE commercial wrapper — for more than a year. What changed at SAPPHIRE 2026, according to Forrester’s analysis, is that SAP moved on the underlying substance of that complaint: a “significant subset” of Joule assistants and agents will now operate on-premises for RISE customers who have already committed to a modernization path, which Forrester calls “the largest concession SAP has made to its installed base in three years” and attributes directly to DSAG pressure. It is a partial answer to DSAG’s ask — access is tied to having started a modernization commitment, not available unconditionally to any on-premises customer — but it is a real move in that direction.

RISE and GROW: the migration path SAP is betting on

SAP paired the technology announcements with an overhaul of its commercial migration programs, on the premise that “simply plugging AI agents into your system landscape will drive zero value” without accompanying change management. RISE with SAP customers now receive a contractual commitment to activate three Joule Assistants within the first year. SAP GROW customers, the public-cloud track aimed largely at mid-market and greenfield implementations, receive more than 20 AI assistants from day one.

SAPinsider reports that the updated offerings include on-site architects and consultants and an AI-powered ERP migration platform targeting up to a 50% reduction in migration effort — SAP’s own target, not an independently verified outcome.

PART 2 · WHAT’S ACTUALLY AVAILABLE

What’s shipping now, and what’s a promise for later

The single most useful thing an independent analyst can do at a keynote-heavy event like SAPPHIRE is separate announcement date from availability date. In May, Forrester’s post-event research note was the most precise public source on this; today, with the quarter nearly closed, it’s worth pairing it with what SAP has since confirmed on its Q2 earnings call, July 23:

SAP AI Agent Hub — two of its six capabilities are already generally available; the remaining four (identity management, observability and performance monitoring) are still targeted for Q3 2026, which closes in a few weeks. Bundled into SAP Business AI Platform at no additional charge.
Autonomous HCM assistants — targeted for June 2026 general availability, per Constellation Research’s reporting.
SAP Enterprise Planning agents (Business Data Cloud + Analytics Cloud) — general availability expected Q3 2026.
Joule Work — mobile experience generally available now; desktop general availability targeted for the second half of 2026; full agent-to-agent (A2A) interoperability targeted for Q4 2026, which Forrester calls “the architectural milestone that actually matters” for multi-vendor agent ecosystems.
SAP Autonomous Suite (Finance, Spend, Supply Chain, HCM, CX) — general availability, in aggregate, is targeted for Q3 2026 or later, with some finance agents already generally available.
Free commercial terms — Joule Studio 2.0 and the Joule agent runtime remain free through December 31, 2026. No 2027 price list has been published, but in July Klein signaled the direction: moving from per-user licensing to pricing tied to agent-delivered value.

Forrester’s blunt May summary is still a useful gut check for anyone tempted to treat the keynote as a done deal: “The vision is credible. The execution is partial. Most of the 224 agents and 51 assistants sit in mixed GA, early-adopter, and preview status.” Three and a half months on, with targets revised upward and the acquisitions closed, that caution still holds — separate the announcement from the actual availability — even as the specific numbers have moved. It’s the distinction that matters when a board asks what SAP promised versus what’s usable today.

What’s happened since the keynote: July results and early production cases

SAP’s Q2 2026 earnings call on July 23 is the first serious public check on what was promised in May, and it surfaces facts that weren’t available then.

Production cases with numbers — Amadeus reports roughly 40,000 reconciled transactions via Autonomous Finance agents; Danish distributor Lemvigh-Müller reports more than 90% touchless purchase-order processing and 98% invoice-matching accuracy.
The RISE reference list grows — alongside names cited in May (Bayer, Novartis, Ericsson), SAP included Spain’s ACCIONA among customers advancing their migration under RISE.
An outcome-based pricing signal — Klein confirmed SAP’s intent to move toward pricing tied to agent-delivered value rather than per-user licenses, though without a date or price list yet — the first public hint of what the 2027 commercial model might look like.

The model layer: why Anthropic’s Claude matters

One detail that got less attention in SAP’s own messaging than it deserves: Forrester’s analysis identifies Anthropic’s Claude as the primary reasoning model behind Joule, alongside SAP’s own SAP-RPT-1 tabular foundation model (which shipped the same day), and partnerships with Mistral and Cohere positioned as sovereign, region-specific alternatives.

Forrester’s read is that this creates a form of ecosystem bifurcation — Microsoft-OpenAI on one side, SAP-Anthropic on the other — with Salesforce, Workday and Oracle remaining comparatively model-neutral. For regulated industries running risk assessments on AI vendor concentration, it is an architectural dependency worth naming: the analysis flags it as “a risk that becomes board-level in regulated industries within 24 months,” contingent on continued partnership stability with Anthropic. The Prior Labs acquisition, flagged in May as pending precisely to reinforce this tabular layer, closed in July.

PART 3 · WHAT ANALYSTS AND THE COMMUNITY SAY

What analysts and the SAP community are saying

Coverage of SAPPHIRE 2026 splits fairly cleanly into three registers: SAP’s own framing (ambitious, forward-looking, delivery-focused), independent analyst research (credible-but-cautious), and practitioner/user-group commentary (focused on commercial terms and access).

The credible-but-cautious view

Forrester titled its research note “SAP Sapphire 2026: The Autonomous Enterprise Is Credible, But It Comes With Concentration Risk,” and its guidance to enterprise technology leaders: “Commit at the architectural pattern level. Pilot at the product level. Define explicit go and no-go criteria before declaring SAP your strategic AI architecture for 2030.” Forrester also references its own broader research finding that 21% of enterprise SaaS decision-makers already cite vendor lock-in as a top commercial concern — a baseline risk that compounds with each layer SAP consolidates under one roof.

Constellation Research’s Larry Dignan raised a similar structural question in his event analysis: every major enterprise software vendor with a suite and a user interface — ServiceNow, Salesforce, Microsoft, Workday, and hyperscalers such as AWS — is chasing the same conversational, agent-orchestrated front door SAP just unveiled with Joule Work. His conclusion is that Autonomous Finance will be the proving ground: “Should SAP’s finance agents deliver well and optimize cash flow, enterprises will have a valid reason to continue with SAP.” If they don’t, the broader autonomous enterprise pitch loses its anchor.

The practitioner view

SAPinsider’s analysis, aimed squarely at SAP practitioners rather than the investor audience, lands on a practical framing question: “Does this agent understand my process, or just my data?” Its recommendation is that governance for non-SAP agents — running in Salesforce, ServiceNow, Microsoft, or homegrown tooling — belongs on every SAP customer’s 2026 planning agenda regardless of whether they adopt SAP AI Agent Hub specifically, because the absence of any cross-vendor governance layer is, in its words, “an audit and compliance gap.”

The user-group view

DSAG’s position, consistent across multiple SAPPHIRE cycles and reiterated around the BAIP rename, is narrower and more commercial: AI capability should not be a lever to push customers toward a specific contract model. Its argument — that SAP has not shown a technical reason why AI functionality needs to be restricted to RISE customers — is one that independent commentators covering the user-group side of SAP, including Diginomica, have found credible in past editions. SAPPHIRE 2026’s hybrid on-premises concession addresses part of that ask; the cost-predictability concern under consumption-based pricing remains open.

THE PRACTICAL TAKEAWAY

What this means for CFOs, CIOs and SAP program owners

Strip away the keynote staging, and SAPPHIRE 2026 leaves SAP customers with a few concrete decisions rather than a single verdict. With the year now heading into its final stretch, it’s worth splitting them across two horizons: what’s left of 2026, and what to prepare for 2027.

Before 2026 closes

Use the free period: Joule Studio 2.0 and the agent runtime are free through December 31, and the four still-missing SAP AI Agent Hub capabilities should reach general availability before the quarter is out.
If you’re on RISE, activate your three contractual Joule Assistants now.
If you’re on ECC or on-premises S/4HANA, confirm whether you’ve started a documented modernization commitment — that’s the gating condition for hybrid AI access.

Heading into 2027

Treat the SAP Autonomous Suite domain-by-domain: Finance is furthest along; Supply Chain and Spend follow; CX and HCM lag behind.
Put SAP AI Agent Hub’s governance scope on the same evaluation list as any other multi-vendor agent-governance project, and watch whether SAP expands hybrid on-premises access beyond customers who’ve already started modernizing.
If you’re risk-averse on AI-vendor concentration, ask your SAP account manager what the Anthropic dependency means for your regulatory environment, and whether 2027 pricing for Joule Studio has been set yet.

The bottom line

SAP did not use SAPPHIRE 2026 to announce a roadmap; it used it to announce a delivery schedule, a governance model, and a commercial structure — and independent analysts covering the event largely agree the underlying architecture is more coherent than SAP’s AI messaging has been in the past two years. Where the credible-but-cautious view, the practitioner view, and the user-group view converge is on a single point: the value of the Autonomous Enterprise, for any given company, will be decided by execution over the next two to four quarters, not by the keynote. Autonomous Finance is the domain SAP itself has staked the most on, and it is the one worth watching first, whether or not you run the rest of the suite.

For SAP leaders, the practical takeaway is not “wait and see” or “adopt everything now.” It is to treat SAP Business AI Platform as an architectural commitment worth evaluating seriously, treat individual Autonomous Suite domains as separate pilot decisions with their own go/no-go criteria, and treat the free commercial terms through December 31 as a real but time-boxed opportunity — not a permanent feature of the platform.