Anthropic Launches Claude Cowork: A First Look at a $100 AI Agent

Anthropic Launches Claude Cowork: A First Look at a $100 AI Agent

Artificial intelligence is advancing quickly, and companies are experimenting with new ways for people to interact with it. One such innovation is Claude Cowork, a dedicated agent mode from Anthropic. The feature is built into the Claude desktop app for macOS and offers users a new level of workflow automation. But as the first hands-on experience showed, the $100-a-month subscription price raises questions about the real value of the product and whether the technology is ready for full-scale use.

Claude Cowork arrives amid a broader push to build AI agents that can carry out tasks autonomously. Companies want to move beyond simple chatbots and offer tools that can analyze information on their own, make decisions and act in a digital environment. Anthropic, one of the leaders in large language model development, has decided to join this race with its own take on AI agents. Claude Cowork is positioned as an assistant that can take over routine or complex tasks, freeing up the user's time for more creative or strategic work. The Claude Desktop app, which hosts the new mode, is meant to become a central hub for working with advanced AI capabilities.

To evaluate Claude Cowork, we ran a practical test. The task was to build a detailed study plan for learning the Go programming language. The choice was not accidental: Go is a popular, in-demand language, and learning it takes a structured approach and an understanding of many aspects, from the basics of syntax to advanced concepts of concurrency and networking.

The expectation was that the AI agent could analyze the available resources on its own, determine the best order of topics and suggest concrete steps for mastering the material. Reality turned out to be less rosy. Despite its claimed capacity for autonomous work, the agent showed a number of significant limitations.

Building the plan required substantial user intervention, corrections and clarifications. The autonomy proved more illusory than real. The agent struggled to maintain context, made logical errors in structuring the material and did not always suggest the most relevant or up-to-date approaches to learning.

That casts doubt on its ability to handle truly complex tasks without constant oversight and direction from a human.

The implications of these first tests are considerable. First, the $100-a-month price looks unjustified at the current level of functionality. Users willing to pay that much expect an AI agent to deliver a high degree of autonomy, accuracy and efficiency, which it has not yet demonstrated.

Second, there are questions about the maturity of AI agent technology as a whole. Whether agents can independently carry out complex workflows, especially in professional settings, remains an open question. Current implementations appear to need significant refinement and optimization before they can become full-fledged tools rather than merely advanced assistants that require constant supervision.

That may slow the adoption of such technologies in business processes, where the stakes are high and mistakes are unacceptable.

In conclusion, the launch of Claude Cowork is an important step for Anthropic in developing AI agents, but the first hands-on experience revealed a number of problems. Despite ambitious claims, the technology has not yet reached a level that would justify the high subscription price. The real autonomy of agents remains in question, and they have a long way to go before they can become reliable helpers in complex workflows. Users should be cautious and not place excessive hopes on the current capabilities of AI agents, especially given their high price.