--- title: "Cowork Is a Big Disappointment: $285 Billion Wiped Because of This?!" description: Claude Cowork triggered a $285 billion SaaS selloff. After using it on Windows, the hype does not match the product. Same model as Claude Code, worse results. author: Darie Nani (Editor-in-Chief) date: 2026-02-19T11:03:54.000Z updated: 2026-02-26T17:55:07.024Z canonical: https://www.sovereignmagazine.com/article/cowork-is-a-big-disappointment-285-billion-wiped-because-of-this image: https://cdn.nanimediahouse.com/claude-cowork.jpg categories: Artificial Intelligence content_type: Opinion region: Global publication: Sovereign Magazine --- Anthropic’s Claude Cowork erased $285 billion in SaaS market capitalisation on 3 February 2026. Thomson Reuters posted its largest single-day drop on record, falling over 15 per cent. RELX shed 14 per cent. LegalZoom dropped nearly 20 per cent. Salesforce, Adobe, ServiceNow and DocuSign lost between 7 and 11 per cent each. Indian IT outsourcers Infosys and Wipro fell over 5 per cent on their ADRs. Nearly $2 trillion in software sector value has evaporated since the start of the year. All of this because of a desktop app that stalls mid-task, ignores instructions and is, put simply, not a good product. ## It Is Broken on Windows Cowork launched on Windows on 10 February 2026, extending the product to roughly 70 per cent of the desktop market. Within days, GitHub filled with bug reports. Users report everything from API connection failures on Windows 11 Home to workspaces becoming permanently unusable during routine tasks to a background Hyper-V VM consuming over 10GB of disk space and regenerating after deletion. These are individual reports, not confirmed defects, but the volume and consistency paint a picture. My own experience lines up. The app is slow. It stops mid-task without explanation. It ignores instructions given moments earlier. It loses track of what it is doing within a single conversation. On more than one occasion it simply stopped responding and had to be restarted. This is not an edge case. It is the baseline experience of using Cowork on Windows in February 2026. ## Same Model, Shockingly Worse Output I have used Claude Code in the terminal and inside VS Code for months. Cowork runs on the same model (Opus 4.6). The difference in output quality is not subtle. Claude Code in the CLI is precise. It reads project files, loads persistent memory from previous sessions, follows a CLAUDE.md configuration file and connects to external tools through MCP servers. It does what you ask, in the order you ask it, and it remembers why. Cowork does not. It confuses file references across documents. It drifts from instructions mid-task. It is not a slightly degraded version of Claude Code. It is a noticeably less reliable product running the same model through a wrapper that actively makes it worse. Even Claude Code embedded inside the Cowork desktop app (which does retain the memory structure) produces worse results than the same session in a terminal or VS Code. Same model. Same files. Same project. The only variable is the wrapper. The fact that a wrapper alone can degrade output this much from the same underlying model is, frankly, embarrassing for Anthropic. ## No Memory, No Continuity, No Point Cowork without Claude Code does not support CLAUDE.md files. It does not retain memory between sessions. Anthropic’s own help centre confirms this as a stated limitation. There is no auto-memory directory. Every conversation starts from zero. The model has no idea what your project is, what conventions you follow or what happened last time you used it. Claude Code solved this months ago. The CLAUDE.md file loads project context at the start of every session. The memory directory carries forward decisions and patterns. MCP servers connect databases, APIs and version control into the same context window. This is what turns a language model into something that can sustain work over time. Cowork strips all of it out and ships the remainder as a productivity tool. The result is a product that cannot maintain a working relationship with anything that lasts longer than one conversation. ## $285bn Selloff Based on Tech Preview Wall Street saw demos. Coverage called Cowork ‘Claude Code for the rest of us.’ Investors priced in the assumption that autonomous AI agents were ready to replace per-seat software at scale. They are not. Cowork is a research preview. Anthropic labels it as such. But the market did not trade it like a research preview. It traded it like a finished product, and $285 billion moved on that assumption. Anyone who actually used Cowork would have seen the stalling, the ignored instructions and the lost context within minutes. The gap between the coverage and the product is the difference between a demo and daily use. Claude Code already proved that current models can do sustained, context-aware work when given the right infrastructure. Cowork proves that removing that infrastructure and wrapping the same model in a buggy desktop app is enough to undo all of it. Maybe the money men should actually try it before panic selling next time. ## Further Context **Q: What is the difference between Claude Code and Claude Cowork?** Claude Code is a terminal-based agentic tool for developers. It reads project files, maintains persistent memory across sessions through CLAUDE.md and auto-memory directories and connects to external tools via MCP servers. Cowork is a graphical interface built into the Claude desktop app aimed at non-developers. It can read and edit files in a designated folder but does not support CLAUDE.md files, has no persistent memory between sessions and offers more limited tool integrations. Both run on the same Opus 4.6 model but differ significantly in the infrastructure surrounding it. **Q: Did Claude Code build Claude Cowork?** Anthropic has confirmed that much of Cowork was built by Claude Code itself, with the development taking approximately 10 days. This makes Cowork both a product and a demonstration of Claude Code’s capabilities. The tool used to build Cowork remains significantly more capable than the product it created, particularly in memory persistence, context management and multi-file handling. **Q: Does Claude Cowork retain memory between sessions?** No. Anthropic’s own help documentation states that Cowork does not retain memory from previous sessions. Each conversation starts without context from prior work. Claude Code, by contrast, maintains an auto-memory directory that accumulates project-specific notes and decisions and loads a CLAUDE.md file with project instructions at the start of every session. This difference in memory architecture is the primary reason Claude Code produces more consistent and context-aware output over time.