|
Simmering anxieties about AI’s threat to subscription software have turned into a raging nervous breakdown on Wall Street in the past week, thanks in part to Anthropic’s recent AI launches (and despite Anthropic's own love for SaaS apps). Even Palantir, which has lately posted blockbuster growth, has been caught up in the selloff. Enter Jensen Huang. Speaking at an event hosted by Cisco this week, Nvidia’s CEO had a simple message: “software is a tool.” The notion that “AI is going to replace” software firms “is the most illogical thing in the world and time will prove itself.” Huang argues AIs themselves will use enterprise apps en masse. But the OpenClaw-ish future that Huang is describing—where AIs run around, using corporate apps the way humans do, except much faster and in the background—does not necessarily mean the biggest system-of-record apps will maintain or accelerate their growth. First off, if human headcount growth slows because of efficiency gains produced by these tool-using AI agents and others, software firms that sell per-seat subscriptions will suffer. Even if headcount doesn’t plateau, Huang’s vision of super useful AIs accessing a variety of apps (tools) to handle white-collar tasks raises questions about why a customer would want to pay for built-in AI features of a specific enterprise app. Wouldn’t they prefer using (and paying for) a super AI agent that can access any and all of the apps simultaneously, from Salesforce to Excel? In that case, the battle will be around the “agentic layer” on top of enterprise apps and the data they store. This is surely one reason tech giants such as Microsoft, Anthropic, OpenAI and Google are developing super AI agents controlled from a computer’s operating system or browser. And in the never-too-early department, firms including Microsoft, Amazon and Salesforce are already trying to play the role of aggregator, pitching their own tools to manage and direct agents made by different companies. OpenAI this morning announced its own efforts in this regard. The individual app owners could try to block super agents from accessing their apps. That wouldn’t be a surprise, if our recent reporting on the AI “data wars” is a guide. But it’s hard to imagine such blocking will last forever, given that customers will likely demand the kind of broad access offered by super agents. We are getting ahead of ourselves, though. As OpenClaw has shown, these super agents aren’t ready for prime time. (See: Microsoft’s recent assessment.) They are a security nightmare, and a lot of scaffolding still needs to be developed for such agents to be trusted with sensitive data or access to live codebases. Today’s AI Agenda has more on the startups developing such scaffolding. Just as it took Waymo a decade and a half to get good enough to be trusted with people’s lives, it may take longer than you think for digital agents to be trusted to not tank a multinational corporation. The stakes are simply too high, and companies can’t afford mistakes. That doesn't mean there won’t be real pain caused by the software selloff, even if it doesn’t exactly reflect SaaS companies’ recent financial performance. But the reality that agents are still far from ready will buy time for the enterprise app makers to make their products (and in-app agents) more useful and to figure out how to get customers to pay more. Perhaps they should borrow some lessons from Palantir, which is cleaning up financially—and trading at a high premium—while other software firms sweat.
|
