In January this year, software firms’ stocks fell off a cliff.
SAP got hit too. Over one trading session, it was down 15%. The steepest drop in five years. Cloud bookings were strong. Revenue beat expectations. None of it mattered. Investors had decided AI might not need an ERP system at all.
In May this year, SAP CEO Christian Klein announced SAP’s Autonomous Enterprise. It was a reaction to what was later called the SaaS Apocalypse.
What Got Announced
Klein announced 50-plus Joule Assistants. Two hundred specialised agents. The big change sat in the small print. Every RISE customer got a contractual commitment. Three Joule Assistants, activated within year one. GROW customers would get the full portfolio from day one.
Contractual. Not “available.” Built into the deal you sign.
That’s not a product launch. It’s a distribution decision.
The Dashboard Problem
Car makers are fighting the same battle right now. It hasn’t gone the way you’d expect.
For years, drivers got Apple CarPlay in nearly every new car. It worked. Drivers loved it. Then automakers worked out what was really happening. Apple was collecting the data: Driving patterns, destinations, habits. The car maker built the steel box. Apple owned the relationship with the person inside it.
GM went furthest, stripping CarPlay from every model by 2028. CEO Mary Barra was blunt about why. GM wants to own the digital relationship with its customers, not rent it back from Cupertino.
“Which would you rather own? The field, or the harvester? SAP is betting the field matters more.
So far, the gamble looks shaky. Roughly four in five buyers call CarPlay a deal-breaker. Most manufacturers such as Toyota, Ford, BMW, Hyundai, simply kept it rather than fight that war.
Nobody has won yet. But the principle hasn’t changed. Whoever owns the interface owns the customer, even when the technology underneath came from somewhere else. Experts call this process dis-intermediation.
SAP is making the same bet as car makers, with better odds. Switching ERP isn’t like switching cars. Nobody trades in S/4HANA because Joule annoyed them for 18 months. The lock-in appears to cut in SAP’s favour in a way it never cut in GM’s.
AI Is A Combine Harvester
Think of AI as a combine harvester. An exquisite machine. It cuts and threshes a field of wheat in a single pass, faster than a hundred people with scythes. But the machine is only as good as the field it’s sent into. Put it somewhere full of stones and crop sown three different ways, and it doesn’t fail gracefully. It jams. It leaves half the harvest on the ground.
Which would you rather own? The field, or the harvester? SAP is betting the field matters more.
Joule doesn’t run on a combine SAP built. It’s an orchestration layer. It routes requests to foundation models from OpenAI, Google, Mistral, and Anthropic, blended with some SAP-specific models trained on things like ABAP code. At Sapphire, SAP went further and named Anthropic’s Claude as the primary reasoning engine behind its newest agents. Most of SAP’s production agents reportedly run on Claude today.
So SAP’s pitch isn’t “our AI beats their AI.” SAP has all but conceded that race. The pitch is that the model isn’t what’s valuable. What’s valuable is the Knowledge Graph. The Company Memory. Decades of clean process and trusted data sitting inside customers’ own SAP landscapes. It’s the layer no hyperscaler can replicate by training a bigger model.
The model is the combine. Anyone can rent one – OpenAI, Anthropic, Google all sell access by the hour. What most enterprises lack is the flat, destoned field. Twenty years of inconsistent customisation. Master data that means something different in Frankfurt than in Singapore. Point the best model in the world at that mess and it still jams.
That’s SAP’s advantage, and a defensible one. And it’s the same argument Pivot makes about standard SAP, for what it’s worth. Clean process and trusted data are what make any layer on top, AI or otherwise, actually work.
A Question CTOs Must Ask
Should CTOs let Joule become the default front door to their business data?
Two questions CTOs must ask.
First: where does your data actually sit, and who can reach it without going through SAP at all? Clean, well-modelled data, accessible through MCP or equivalent standards, gives you options. Any model can reach it, whether or not Joule’s in the room. Data locked inside SAP-specific structures gives you fewer.
Second: who’s accountable if Joule gets it wrong? Three assistants, contractually activated in year one, means three live points where an AI agent is acting inside your finance or supply chain processes from month one. Someone needs to own that risk before go-live, not during it.
An Answer
Christian Klein is betting that owning the field beats owning the combine. Anyone can lease the best harvester on the market. Not everyone can hand it a field worth harvesting.
The dashboard war suggests that bet isn’t automatically a winning one. Owning the interface only protects you if customers can’t easily leave for somewhere the experience just works. GM is finding that out in real time. Drivers who hate GM’s native software can switch to a different car.
But a CTO’s situation is less binary than that. Leaving SAP entirely is a multi-year decision few CTOs will make over an AI assistant. But you don’t have to leave SAP to decline Joule. Stay on S/4HANA, keep the RISE contract, and point Claude, or any model with the right access, straight at your own business data over MCP. Joule becomes optional. The ERP underneath doesn’t.
Does Joule offer anything Claude alone doesn’t?
Claude reaching into SAP’s data through MCP gets you the reasoning and the access. It doesn’t get you the Scenario Catalogue SAP has spent two decades building. That’s the mapping between “close the books” and the 40 transactions that involves, with the role permissions, approval chains, and audit trail already wired in. That orchestration layer is real engineering, and right now Joule is the only place SAP ships it pre-built. Build it yourself with raw model access and you’re recreating, line by line, the workflow logic SAP already encoded.
Whether that’s worth the lock-in depends on one thing. How much of that orchestration you could build or buy elsewhere, faster or cheaper, without SAP’s permission. For most CTOs today, building it elsewhere is still slower than activating what’s contractually already in the RISE deal. That gap closes every time Model Context Protocol-style standards mature. It is not closed yet
What I will say is this: get your field flat and de-stoned. Get your data and processes standard, clean, and yours to move. That way, you’ll have a choice about who harvests it. Joule, Claude, or whatever comes next. The contract shouldn’t be the thing making that decision for you.
Image courtesy of SAP.
