Unlock the Editor’s Digest for free
Roula Khalaf, Editor of the FT, selects her favourite stories in this weekly newsletter.
The rapid advances in artificial intelligence since the launch of ChatGPT have taken place deep inside the leading AI labs, out of sight to most people. By contrast, the stock market lurches that come as investors wake up to the disruptive potential of all this new technology can be all too visible.
The devastation of software stocks over the past month is a case in point. AI agents — software tools that can take actions on behalf of a user — have long been talked about in the AI world. The market ructions are a sign that they may finally be at hand.
The main source of investor angst has been a move by Anthropic and OpenAI to repurpose their code-generating tools as general-purpose agents, capable of carrying out a wide range of actions for non-technical workers. Need to clean up your email inbox and organise your expenses receipts, or handle more specialised jobs like producing marketing copy or analysing a legal contract? You may soon be able to ask an all-purpose agent to do all this and more, without having to open all the different apps currently used to accomplish these tasks.
This presents a severe risk that established software companies will be usurped. Rather than replacing existing software, automated agents could become a new layer on top, a primary computing interface for many workers and a chokepoint for AI firms to claim a bigger slice of corporate IT budgets.
Wall Street has woken up to this as a threat in particular to software-as-a-service, or SaaS, companies, which use software to provide corporations with a wide range of services, from human relations management to handling customers. However, many companies outside the tech industry also rely on software to shape the services they sell, potentially putting them in the line of fire as well. Business-to-business data providers and wealth management companies have been among those caught up in the recent stock market carnage, but they will not be the last.
Investors did not stop to ask too many questions before concluding that this was a threat to much of the SaaS industry. In reality, the picture is more mixed. Most exposed are companies whose software is used for non-essential work processes that can easily be replicated by others. There are no moats to protect businesses like these as AI agents become ever more capable and step between software and users.
Software companies that act as systems of record, holding important corporate data, are more deeply entrenched. Such companies also embed core business processes that are seldom changed. Even if they look elsewhere for new AI capabilities, customers are unlikely to rip out these essential systems. But there is still a risk that these software businesses will fade into the background, relegated to the role of essential but largely ignored utilities that miss out on the growth from AI.
That does not mean some of these companies won’t succeed in reinventing themselves around AI. It takes time to build enterprise-grade systems, to persuade customers to test and use them, and to retrain workers. The AI insurgents will find this out for themselves when they seek to turn their new agents from eye-catching demos into revenue-generating products. High among customer concerns will be how to defend against the heightened risk of prompt injection attacks, where the new agents are misused to reveal sensitive data or take unauthorised actions.
The message from the financial markets, however, is that SaaS companies need to move much faster. Incumbents in other industries who think time is on their side as they figure out a response to the AI threat should also take note. The stock market’s AI reckoning, when it comes, can be swift and brutal.