Like most people today, I have invested a decent amount of my savings in the stock market. On the 4th of February 2026, I was reading the news and checking prices as I normally do, when something unusual caught my attention. A range of DAX companies — Siemens, BASF, Deutsche Telekom — were swinging between -6% and +6% intraday. These are not speculative tech stocks. The volatility of DAX constituents is typically contained, driven by fundamentals and slow-moving macro signals. Something had changed in the anticipation of the crowd. I wanted to understand what.

After some research, the thread led me to a product release from Anthropic: Claude Cowork. Several news agencies were already drawing comparisons to the so-called DeepSeek moment — the day a Chinese AI company published a large language model at a fraction of the cost that Western incumbents had assumed necessary, triggering a sharp reassessment of AI investment theses across global markets. The market reaction to Claude Cowork carried a similar character: not excitement, but recalibration. Investors were not buying into a new trend — they were selling out of an old one.

To understand why, it helps to step back and trace the arc of where we are. Most people became aware of large language models through ChatGPT — the first application to make AI tangible and experienceable to a broad public through a simple chat interface. Anthropic was founded by former members of OpenAI and has, from the outset, positioned itself at a different layer of ambition. Where others focused on the chat paradigm, Anthropic has consistently oriented its development around agents: software that does not merely respond, but acts — performing tasks and executing workflows autonomously, without requiring a human in the loop for every step.

The distinction matters more than it might seem. An AI chat answers questions. An agent manages a process. And the two primary obstacles to making agents broadly useful have always been the same: how do you give an agent access to the right tools, and how do you give it access to the right knowledge? Anthropic has addressed both systematically. The Model Context Protocol — MCP — standardises how agents integrate with external tools and systems. Agent Skills standardise how specialised knowledge and capabilities are packaged and shared. Taken together, these two definitions form the architectural foundation for a marketplace — one that is now opening up, and one that Anthropic does not need to fill itself, because the crowd will.

This is the crowdsourcing strategy at the heart of Anthropic’s position: not to build every capability, but to define the standards through which capability can be contributed, composed, and discovered.

The analogy to Docker is instructive here. Docker did not win by providing the best containers — it won by becoming the standard through which containers were defined, distributed, and trusted. Once that standard was adopted, a hub emerged, and the hub compounded in value with every new image contributed. Anthropic appears to be making the same bet at the level of agent behaviour. If MCP and Agent Skills become the lingua franca of workflow automation — which the pace of adoption among enterprises and developers suggests is already underway — then the value of the ecosystem will accumulate around the standard-setter, regardless of which individual agents become dominant.

What does this mean for the traditional software industry? The stock market, as it often does, reacted before the analysis was complete. But the underlying logic is coherent. Enterprise software companies have historically derived significant value from the friction of integration — from the cost and complexity of connecting their platforms to the broader ecosystem. Agents, particularly when they operate through open standards, reduce that friction structurally. If a skilled agent can interact with an ERP system through an MCP connector maintained by the open-source community, the proprietary integration layer of that ERP vendor becomes thinner. The market appears to be pricing this in.

In the longer term, I expect enterprises to begin building private skill repositories — curated libraries of agent capabilities tuned to their processes, their data, and their regulatory context. This is not a distant scenario; it is the natural evolution of what is already happening with internal knowledge bases and process documentation. The difference is that skills, unlike documents, can act. And when they can act, the question of who defines their structure — and who audits, governs, and distributes them — becomes a strategic question of the first order.

The 6% swing in the DAX told me something that a press release could not: that the crowd, in aggregate, has begun to update its model of the future. The question worth sitting with is not whether workflow automation will come — it will — but whether organisations are building the governance and the capability to shape it on their own terms, rather than simply adopting whatever standard emerges from outside. The standards being set today will be difficult to reverse tomorrow. That is worth paying attention to, whether or not you own stocks.