Just recently, two of the most influential voices in enterprise software described the same future of ERPs from opposite sides of the table.
McKinsey published "The end of ERP as we know it?" — a sweeping essay arguing that AI will redefine ERP from the ground up. Days later, at Sapphire 2026 in Orlando, SAP CEO Christian Klein took the stage and unveiled the SAP Business AI Platform: a connected set of products covering data, intelligence, AI agents, and a new way for people to interact with the system.
I put the two side by side to answer a question: if your CIO reads the McKinsey piece tonight, did SAP just deliver what it describes?
The short answer is: mostly yes — with two important gaps and one underrated surprise.
What each side actually said McKinsey describes five things an AI-native ERP needs to have.
A trusted core where business data lives. A semantic layer on top that helps software understand what that data means. A set of AI agents that act on behalf of users. A new way for people to interact with the system through intent rather than screens. And, sitting above all of it, a continuous measurement layer that tracks whether the AI is actually improving the P&L.
SAP showed up at Sapphire with a product for each of these — except the last one.
In SAP's language: the clean S/4HANA core stays as the system of record. The new Business Data Cloud pulls relevant data into a single fabric. The Knowledge Graph makes the business comprehensible to AI. Joule Agents execute the work. Joule Work replaces screen navigation with intent-driven interaction. The missing piece is the layer that measures whether any of it is paying off — and we'll come back to that.
Where SAP genuinely delivered A single, modern data foundation. This is the most mature part of the SAP story. The Business Data Cloud finally consolidates what used to be a mess of platforms. The migration story toward a clean S/4HANA core hasn't changed — but the surrounding data fabric is now one conversation instead of four. For program leads, this matters: the platform discussion with your CIO is shorter than it was a year ago.
A pre-built map of the business. McKinsey describes a "semantic layer" that explains the business to the AI. The Knowledge Graph is SAP's version, and it ships pre-populated with hundreds of thousands of business objects — tables, data fields, processes, applications. The important word is pre-populated. McKinsey describes this as something each customer would have to build; SAP is delivering it as a starting point. That changes the time horizon dramatically.
Real agents, in production, with names. Roughly 40 ready-to-use AI agents across Finance, HR, Procurement, Supply Chain, and Sales — Cash Management, Dispute Resolution, Accounts Receivable, Expense Reports, HR Service, Production Planning, Maintenance Planning, and more. Each one has a release date. Each one has claimed productivity numbers. The keynote case studies were specific: LC Waikiki cited 70% efficiency gains and 50% error reduction; JPMorganChase ran a live general-ledger migration on stage; Sony reduced certain developer tasks from 3–4 days to 10–15 minutes. The catalog itself — concrete, dated, customer-validated — is the move that anyone working from the McKinsey article alone will not have priced in.
Agents now included, not negotiated. RISE with SAP customers get three Joule Assistants in their first year at no additional charge. GROW customers get the full portfolio at onboarding. For practitioners this is a meaningful shift: the AI capabilities are no longer something to lobby for in procurement. They are on the doorstep.
What McKinsey didn't see coming This is the part of the comparison most likely to be undervalued - and most relevant to anyone actually running SAP programs.
McKinsey writes about AI as if it's essentially the same thing as ChatGPT-style language models, only embedded into ERP. SAP built something more deliberate: a system that uses different kinds of intelligence for different kinds of work, with a named, purpose-built product behind each one.
For business numbers and structured data — the kind that lives in tables — SAP built SAP-RPT-1.5, a model designed specifically for analyzing and predicting from tabular business data rather than for writing prose. SAP is accelerating this line of work with the pending ~$1.17 billion acquisition of Prior Labs, a specialist in tabular AI. For ERP workloads, where almost every decision starts with a table of numbers, this is the right tool for the job — and a much better one than a general-purpose language model.
For ABAP code — the language SAP applications are written in — there's SAP-ABAP-1, generally available since January 2026 and trained on roughly 250 million lines of real ABAP code. As of today, this is the only commercially deployed ABAP foundation model anywhere. For SAP shops with large custom-code estates, that's a productivity event in its own right, independent of any greenfield AI strategy.
For conversation and natural language, SAP routes work through a multi-model hub: Claude as the primary reasoning model, with GPT, Gemini, and Llama also available — and, importantly, Mistral and Cohere as sovereign options, meaning European, regulated, and public-sector customers can keep both their data and their AI provider inside the right jurisdiction.
For searching across unstructured content — contracts, manuals, emails, internal documents — AI needs a specialized kind of database called a vector store. Most companies adopting AI today are running a separate system for this (Pinecone, Weaviate, or Chroma are the common names). SAP built this capability directly into the database as the HANA Cloud Vector Engine, so the same security rules that protect your data in HANA also protect the AI's document searches — with no separate system to manage, govern, or pay for.
This is a quietly significant architectural bet. Most AI conversations today assume that one large language model can do everything. SAP is betting that business data, SAP code, conversation, and document search are each best handled by a different, specialized tool — and is delivering them as one connected stack. Competitors will find this hard to copy in less than a few years.
Where SAP genuinely fell short Three gaps stand out — not because the industry has solved them anywhere else, but because McKinsey calls them out explicitly and your stakeholders will ask about them.
Gap 1 — Measuring whether AI actually pays off. This is McKinsey's top priority, and SAP didn't deliver a product for it. The argument is straightforward: as agents start making real decisions, leaders need to know which ones are moving the P&L and which ones aren't. Without that, AI proliferates without ROI. The ingredients for this exist at SAP - SAP Analytics Cloud, Signavio for process intelligence, and the telemetry inside Joule - but they're not assembled into a single named offering that does the job. If your CFO asks which agents are actually moving EBIT, SAP doesn't have an answer today. This is the most defensible piece of criticism in the entire analysis, and where CIOs are already struggling during the S/4HANA Transformations.
Gap 2 — The AI commercial model isn't simple. McKinsey is explicit that customers need to easily forecast and track AI costs. SAP's pricing is the opposite. AI consumption is metered in capacity units that convert into different rates depending on which underlying model is used, with margin added on top of the provider's price. RISE bundles include some AI consumption — but for active deployments, those bundles deplete in months, not years. A CIO modeling AI spend three years out still has to forecast token usage by role across a non-trivial conversion table. This will get cleaner over time. In 2026, it remains a CFO objection at every steering committee.
Gap 3 — Silence on the human side. McKinsey is clear: as AI takes over more of the design, build, and testing work, the bottleneck moves to change management — whether end users come along for the ride. SAP at Sapphire didn't announce a dedicated AI-enabled change management product. SAP Enable Now and Joule's in-app help are the implicit answers, but they're not new and they're not designed for an AI-native world.
The bottom line One. SAP delivered the architecture McKinsey was describing. Four of the five pieces — data foundation, business map, agents, and intent-driven interaction — exist as shipping or near-shipping products. Anyone arguing that SAP "missed the moment" hasn't read the catalog.
Two. The real gap is on measuring value. McKinsey put this at the top of the stack. SAP put nothing there. Customers will need this within twelve months and will likely get it from a partner or a future acquisition before they get it from SAP itself.
Three. The most underrated part of SAP's strategy is the one McKinsey didn't anticipate at all — purpose-built intelligence for different kinds of work, combined with sovereign options, master data, and a certified AI management system. Those choices are exactly the ones that protect SAP customers from the worst failure modes of AI projects, and they are exactly the choices a competitor will find hardest to copy.
McKinsey described a five-piece architecture in May 2026. SAP just delivered four of them — with one missing layer and one quiet but significant surprise.
That's the conversation worth having with your CIO this quarter.