How German philosophy can beat the SaaSpocalypse
13/05/2026 14:51
The author is a Reuters Breakingviews columnist. The opinions expressed are her own. By Jennifer Johnson LONDON, May 14 (Reuters Breakingviews) - SaaS stocks are still menaced by the spectre of artificial intelligence. The likes of $200 billion SAP SAPG.DE and $137 billion Salesforce CRM.N – who’ve long sold software subscriptions to enterprise clients – are down by roughly a third in the past year. Investors aren’t certain that these companies will be the ones to benefit from the introduction of AI agents to the workplace. To show the naysayers that they aren’t simply yesterday’s databases, the Software-as-a-Service mavens need to get up to speed with a word more typically associated with German philosophers of yesteryear, like Immanuel Kant and Martin Heidegger: “ontology”. In the race to create sticky enterprise AI products, SaaS groups and frontier artificial intelligence labs each want what the other has. Products offered by the likes of Salesforce, SAP and ServiceNow NOW.N are already embedded at the heart of most large organisations – managing their supply chains, finance and customer service functions. But they haven’t quite figured out how to make large language models (LLMs) genuinely useful for their customers, or particularly profitable for themselves. Meanwhile, OpenAI and Anthropic have developed cutting-edge models, but they still need to make them fit seamlessly into day-to-day corporate functions. Both sides are ultimately vying to control the interface where employees ask AI to do work – and where AI systems coordinate actions. That’s where the ontology comes in. In its original context beloved of Heidegger and co, the term refers to the study of being. More recently, however, it has been appropriated by Palantir Technologies PLTR.O and its boss Alex Karp, who as it happens has a doctorate in neoclassical social theory from Goethe University Frankfurt. To the $325 billion defence tech group, an ontology is a living map of the customer’s organisation that its Foundry platform uses to connect raw data to real-world objects, like warehouses and machines. Without these guidelines, an agent won’t be able to understand the significance of stored data, or how to act on it. For instance, it might be able to flag that a crucial shipment is delayed – but not explain what that means for quarterly revenue and manufacturing output. Palantir’s ontology is arguably the engine behind a lofty valuation that, despite recent falls, remains around 35 times’ expected sales in the next 12 months. Given SAP and peers trade on an average of only 4 times forward sales, the market doesn't seem to be expecting much in the way of AI-ready innovation. In fact, investors have interpreted the release of enterprise-focused AI products as a direct challenge to software’s prevailing business model. The fear is that agents like Anthropic’s Claude Cowork will be able to execute tasks that are today performed by human workers at a computer. This means that firms like SAP could end up selling fewer seats, or licences, to use their products – while once-satisfied customers may question why they’re paying so much to begin with. However, an agent can’t simply be given an employee login and let loose on a company’s IT network – it needs to understand the connections between disparate datasets stored across different software systems. According to Bain, SaaS companies can convert human labour into software spending by automating this kind of coordination work. The management consultancy reckons there’s a $100 billion market for so-called “cross workflow” agents – with 90% of it uncaptured. There are AI-native startups, like $15 billion customer-service platform Sierra, playing in this space. Meanwhile, OpenAI and Anthropic are partnering with investors to build services firms that embed their models in enterprise workflows. In other words, there’s a kind of usefulness arms race going on in the world of agentic AI. The fact that legacy SaaS players already store clients’ data, and understand their business processes, should give them a fighting chance. This week, SAP said it was partnering with Anthropic to put the lab's Claude model inside its own AI tools, meaning its agents will be able to carry out tasks across its suite of enterprise software. In this arrangement, AI agents become a feature of the SAP environment, not a separate layer customers seek out and bolt on themselves. But to re-rate, the incumbents will also need to show that their new AI offerings and partnerships are driving top and bottom-line growth. There are positive signs at Salesforce, where the Agentforce platform brought in $800 million in annual recurring revenue in the three months to the end of January. Though that’s a 169% year-over-year increase, it only accounts for about 2% of Salesforce’s $11 billion in quarterly revenue on an annualised basis. On a first-quarter earnings call last month SAP CEO Christian Klein told analysts that the company still had work to do on its own ontology layer. The remark suggests the company knows its hugely popular enterprise resource planning (ERP) systems contain masses of valuable data, but that it’s not yet intelligible to AI agents. If the SaaS companies can’t find a way to demystify the data they store, they may find themselves embracing another of Heidegger’s famous concepts: “being-towards-death”. Follow Jennifer Johnson on Bluesky and LinkedIn. Palantir commands a stratospheric valuation next to SaaS incumbents https://www.reuters.com/graphics/BRV-BRV/myvmyrxrovr/chart.png Investors see new business-focused AI agents as bad news for SaaS stocks https://www.reuters.com/graphics/BRV-BRV/gkplkeaygvb/chart.png (Editing by George Hay; Production by Streisand Neto) ((For previous columns by the author, Reuters customers can click on JOHNSON/Jennifer.Johnson@thomsonreuters.com))