Artificial intelligence has quietly embedded itself into everyday life. It drafts emails, plans holidays, writes social-media posts and explains unfamiliar ideas. For most people, this has felt like a steady but manageable productivity upgrade. Yet recent events suggest something more consequential is under way ” a shift that has unsettled investors, executives and policymakers alike.
The catalyst was not a recession, a geopolitical shock or a credit event. It was a technology demonstration.
Rapid growth
In recent weeks, Anthropic, one of the most closely watched AI firms, showcased advances that underline how rapidly artificial intelligence is moving beyond its original role as a digital assistant. These systems are no longer confined to answering prompts or accelerating discrete tasks. They are beginning to assume responsibility for entire workflows ” planning, executing, reviewing and refining work with minimal human intervention.
That distinction helps explain why markets reacted so sharply. Shares of several listed software-as-a-service (SaaS) companies fell abruptly following the demonstration, erasing tens of billions of dollars in market value over a matter of days. Investors appeared to be repricing not earnings, but relevance.
SaaS slowdown
For more than a decade, SaaS firms prospered by selling tools that helped humans work faster. Customer-relationship systems, legal databases, marketing dashboards and analytics platforms all assumed a human operator. Software supported decision-making; people remained firmly in control. Even earlier generations of AI fitted neatly into this model, offering suggestions rather than outcomes.
That assumption is now being tested.
Anthropic and its peers are increasingly focused on building systems that can handle extended tasks end-to-end. Recent advances in long-context reasoning ” allowing models to process entire codebases, legal archives or financial histories in a single reasoning flow ” have reduced one of AI’s most persistent weaknesses:
fragmentation. More importantly, research labs are experimenting with what are often described as “multi-agent” systems, in which several AI instances collaborate on a problem, each performing a distinct role.
Multi agent systems
The result begins to resemble a digital organisation rather than a tool.
In software development, such systems can already draft, review and test code across large projects. In legal and compliance work, they can sift through vast bodies of precedent, draft contracts and flag risks. In marketing, they can plan campaigns, generate content, test variations and refine strategy continuously. In finance and operations, they can reconcile accounts, identify anomalies and optimise processes in near real time.
Autonomous workflow
What unnerves investors is not that these capabilities exist, but that they threaten the economic logic of much of the software industry. If an AI system can perform an entire workflow autonomously, many standalone tools become redundant. The software itself may still function, but its value collapses when execution is bundled into a single intelligent layer. Why pay for several subscriptions when one system can coordinate across them all?
This is the logic behind what some have dubbed a “SaaS reckoning”. It does not imply the disappearance of software firms, but rather a painful consolidation. Products designed for human operators may struggle if machines become the primary users ” or if they are bypassed entirely.
Importantly, this is not simply a story of machines replacing people. Instead, AI is eroding the software layer that humans have depended on to do their jobs. Roles centred on operating tools are most exposed. Roles involving judgment, accountability, creativity and strategic oversight remain harder to automate, though even these will be reshaped.
An uneven transition
The transition will be uneven. In the United States, firms are already experimenting aggressively, aided by relatively light regulation and a culture of rapid adoption. In Europe, where rules around data protection, liability and algorithmic accountability are stricter, deployment is likely to be slower and more cautious. Elsewhere ” from India to parts of Africa ” the impact may be felt less through job displacement than through sudden productivity gains that allow smaller firms to compete globally.
This divergence matters. It suggests that AI’s economic shock will not arrive everywhere at once, nor look the same in every market. Policymakers who assume a single global trajectory may find themselves unprepared.
AI as a colleague
Anthropic is careful to frame its work in terms of safety and alignment rather than artificial general intelligence. Yet the direction of travel is clear. Longer memory, self-correction, autonomous execution and coordinated agents all narrow the gap between today’s AI and systems capable of handling much broader forms of intellectual labour.
That is why a product demonstration ” not a macroeconomic event ” was sufficient to unsettle markets. It made visible a future that many suspected but had not yet fully priced in.
The era of software as a passive tool is ending. What replaces it will look less like an app and more like a colleague ” one that never sleeps, never forgets and increasingly, never asks for permission.
