10 Days vs. 10 Years: What Claude Cowork Teaches Us About Building the Right Tool

Introduction

Generally, the software products require months or years to be developed. The teams invest a lot of time in writing code, fixing bugs, iterating on features, and hoping that their product is really what the users want.

Claude Cowork was created in 10 days. That was not by a huge engineering team working around the clock, but by an AI self-developing.

Even more astonishing: it came into being because users were already “misusing” Claude Code, taking a developer tool to plan vacations, recover wedding photos, and control an oven. Anthropic didn’t spend years figuring out the ideal productivity tool. They saw what people actually did, then built that product in a week and a half.

This is not only the story of how fast. It is about the major change in the way we develop software: instead of thinking up what the users might need, we watch what they are already doing and help them get on with it.

The Unexpected Origin of Claude Cowork

Anthropic, the maker of Claude, anticipated that developers would leverage their newly launched product, Claude Code, for a single purpose: coding. Writing functions, debugging, refactoring projects, and essentially developer work. However, people turned to it for quite a lot of other things.

Here is the list of things that users did with Claude Code instead:

  1. Planning vacations and researching destinations.
  2. Building presentation slide decks.
  3. Cleaning up email inboxes and cancelling subscriptions.
  4. Recovering wedding photos from corrupted hard drives.
  5. Monitoring the growth of tomato plants.
  6. Operating smart ovens.
  7. Doing taxes.
  8. Designing knitting patterns.
  9. Analyzing genome data.

Boris Cherny, head of Claude Code at Anthropic, recognized the trend: “Ever since we launched Claude Code, we have observed people using it for various non-coding tasks.”

The indication was loud and clear: Claude Code was not merely a developer tool. It was a general AI assistant capable of performing any task, but hidden in the CLI (command-line interface).

Most companies would have taken years to “properly” create a general productivity tool. Anthropic did it in 10 days.

How Claude Built Claude (Yes, Really)?

This is where the plot turns crazy:

  • The team didn’t build Claude Cowork. It was Claude Code that built Claude Cowork.
  • Boris Cherny confirmed on social media: “The whole product’s code was written by Claude Code.”

The timeline:

  1. Day 1 – 2: The team decided on what Cowork should be.
  2. Days 3 – 10: Claude Code wrote the whole app.
  3. Day 10: Released to production as a research preview.

Felix Rieseberg, the lead engineer, said it was “a sprint and a half”, about 1.5 weeks. The team had tried out some prototypes before, and Claude Code gave the framework, but the speed is still unheard of.

Here is what it implies:

  • Traditional software cycles: 3-12 months.
  • AI-powered development: 10 days.
  • Speed increase: 10-30x faster.

We’re witnessing AI tools creating themselves. The recursive loop that everybody is sure was at least 3 years away? It’s already here.

Why Speed Matters (But Not How You Think)?

It’s not the fast 10-day timeline that makes it so remarkable. The beautiful part is that they delivered to users exactly what users were already figuring out how to do on their own.

Traditional development usually goes like this:

  1. Brainstorm features (weeks).
  2. Design mockups (weeks).
  3. Write specifications (weeks).
  4. Build (months)
  5. Test (weeks).
  6. Launch (maybe users want it).

Anthropic’s method:

Look at how users are already exploiting existing tools, simplify and smooth out what they are already trying to do, and deliver within 10 days.

This is totally different from the “build it, and they will come” scenario. It is more of a “they’re already here, just make it easier” situation.

What Sets Apart Claude Cowork?

Claude Cowork is not one of those chatbots. It is not an assistant to whom you ask questions. It is an AI agent that literally has file system access and can accomplish tasks autonomously.

Traditional AI assistants:

  1. You: “Write an email about this project.”
  2. AI: “Here is a draft: [text].”
  3. You: Copy, paste, send.

Claude Cowork:

  1. You: “Help me organize my downloads folder by file type and date.”
  2. AI: Actually does the reorganization of your files.
  3. You: Nothing. It is done.

Genuine potential:

  • Claude can read, edit, and create files in specified folders.
  • Claude can make a spreadsheet out of the screenshot of a receipt.
  • Claude can draft a report from unorganized notes.
  • Claude can reorganize a cluttered file system.
  • Claude can extract data and make a structured document.
  • Claude can surf the web when used together with Claude in Chrome.

The chief distinction: Cowork is endowed with agency. It is a self-planner, thinker, and doer.

The Real Innovation: Bottom-Up Development

Most of the firms usually take a top-down approach in developing AI products, for instance:

  1. They first agree on AI assistants, say helping in productivity.
  2. They build the feature/s they think users need.
  3. They launch and hope.

However, Anthropic went bottom up:

  • First, they built a really good coding agent (Claude Code).
  • Next, they observed the users hacking it for non-coding tasks.
  • After that, they removed the command-line barrier.
  • They delivered what the users were already demanding.

This upends the usual innovation model. Rather than pushing features on users, Anthropic drew features out of user behavior.

Simon Willison, Programmer based in the UK: “This is a general agent that looks well-positioned to bring the wildly powerful capabilities of Claude Code to a wider audience. If Gemini and OpenAI do not follow suit, I would be very surprised.”

The competitive advantage is not speed; it is rather starting with real demand proven by the market instead of needs based on hypotheses.

The Market Impact Nobody Saw Coming

Claude Cowork came in like a bomb in the productivity software market.

By July 2025, Claude Code had enticed 115, 000 developers who were processing 195 million lines of code weekly. The venture capitalist Deedy Das reckoned annualized revenue potential at around $130 million.

But here’s the danger to the existing startups: Claude Cowork is capable of doing what hundreds of AI productivity startups have promised but not delivered.

Startups that are now faced with direct competition:

  1. Tools for file organization.
  2. Services for document generation.
  3. Platforms for data extraction.
  4. Apps for workflow automation.
  5. Email management tools.
  6. Photo recovery services.
  7. All of these had raised millions in funding.

All of them are now being pitted against a tool made in 10 days that is included in Claude Max subscriptions ($100-200/month).

The worry for startups: What if the basic AI labs can come up with similar features in 10 days and include them in their main product? Then what will be the defensible moat of the specialized apps that have been built on top of those same models?

The Trade-offs: Speed vs. Safety

Shipping in 10 days means making some compromises. For example, Claude Cowork was launched initially as a “research preview,” and the team was aware of many rough edges.

Known issues:

  1. macOS only (Windows support planned but not scheduled).
  2. Users are experiencing scary error messages.
  3. Problems with calendar connections.
  4. Prompt injection vulnerabilities.

Anthropic admits that there are still some risks: “We’ve built sophisticated defenses against prompt injections, but agent safety, that is, securing Claude’s real-world actions, is still very much a work in progress in the industry.”

The prompt injection threat: If Claude is browsing the web and comes across a piece of malicious content, that content could be used to manipulate his actions. In case Claude is reading a file or website that is compromised, the attackers could issue commands to change Claude’s behavior.

Security researchers express their concern: If an AI system can create its own successor in 10 days, human teams will be up against “an impossible race” to keep up with the audit of what is being produced.

Anthropic’s way of doing things: Deliver the product quickly, learn from the real usage, and make iterative changes swiftly. That is modern SaaS development, except that the AI is developing another AI.

The issue is not if this is risky; it is, of course. The question is whether a careful, slow development approach would be safer or simply slower.

What This Means for Product Development?

The Claude Cowork example teaches us several important lessons for product creation in the age of AI:

Lesson 1: Look at What Users Hack Together – The best product ideas result from observing how users get creative with your tools and use them in ways you didn’t initially intend. They are basically telling you what they really need.

Example: People turned Claude Code (a developer’s terminal tool) into a vacation planner. This isn’t a bug; it’s a feature request shouted at you.

Lesson 2: Remove Barriers, Don’t Add Features – Cowork and Claude Code are essentially the same tools. It’s just that Cowork has done away with the command-line barrier.

Only sometimes, a new product is really just your existing product but with fewer hurdles.

Lesson 3: AI Speeds Up Execution of Already Validated Ideas – It took 10 days to make Cowork because the fundamental idea was already there. Claude Code was working; Cowork only made it more user-friendly.

AI doesn’t accelerate the process of guessing what users want. It accelerates the process of creating what you already know users want.

Lesson 4: Ship, Learn, Iterate – The launch of Claude Cowork was a bit rough around the edges. Anthropic is very transparent about it being a research preview. Users get that.

The other option, spending a whole year making it perfect, means not learning from real usage at all, while competitors are shipping.

Real Use Cases from Early Adopters

Let’s see what people have actually been doing with Claude Cowork:

  • Content Creator Workflow: Rachitsky’s transcript analysis provided a great example of how Cowork can take messy, unstructured data (interview transcripts, notes, research) and give out structured, formatted input (blog posts, summaries, organized files).
  • Small Business Owner: Previously, organizing receipts, extracting expense data, and generating spreadsheets for accounting were considered to be manual data entry tasks or hiring a person to do it. With Cowork, however, such things can be done completely automatically now.
  • Researcher: Keeping track of research notes, PDFs, and web clippings that are scattered around. Cowork produces structured literature reviews and citation databases.
  • Event Planner: Bringing together all files from multiple vendors, making checklists, and budget tracking, all done through the chaos of email attachments and downloaded files.
  • The pattern: Cowork is really good at taking the disorder (scattered files, screenshots, notes) and making it into order (organized folders, spreadsheets, documents).

The Limitations Nobody Talks About

Claude Cowork isn’t a magic wand. There are definite limitations that are worth knowing:

Current Constraints:

  1. macOS only (thus excluding more than 75% of computer users).
  2. Requires a Claude Max subscription ($100, 200/month).
  3. Only limited to designated folder access.
  4. Web access requires separate Chrome integration.
  5. Cannot handle tasks requiring human judgment.
  6. Struggles with highly specialized domain knowledge.

Performance Data:

According to the Center for AI Safety’s Remote Labor Index, Opus 4.5 (the engine of Cowork) managed to finish only 9 out of 240 human freelancer projects that include tasks such as architectural plans and video game development.

In other words, Cowork is brilliant at processing structured, repeatable tasks. Highly intricate creative tasks requiring profound expertise? Still not.

Comparing Claude Cowork to Alternatives

vs. Microsoft Copilot:

  1. Copilot: Deployed in 90%+ of Fortune 500, enterprise-focused.
  2. Cowork: Consumer-first, research preview, limited availability.
  3. Advantage: Microsoft has distribution.
  4. Anthropic has proven agent capabilities.

vs. ChatGPT + Custom GPTs:

  • ChatGPT: Conversational, no direct file access.
  • Cowork: File system access, autonomous task completion.
  • Advantage: Cowork actually manipulates files vs. generating instructions.

vs. Specialized Productivity Apps:

  1. Apps: Purpose-built for specific tasks, polished UX.
  2. Cowork: General-purpose, rough edges, broader capability.
  3. Advantage: Apps are focused; Cowork is flexible.

The Future:What Comes After 10 Days?

If AI is capable of creating production software in 10 days, then what are the world’s possibilities in 12 months?

Short-term predictions (6, 12 months):

  1. Windows and Linux support for Cowork.
  2. Expanded file type handling and automation.
  3. Deeper integrations with existing productivity tools.
  4. Reduced pricing tiers for broader access.

Medium-term (1- 2 years):

  • Multi-agent collaboration (specialized agents working together).
  • Enterprise deployment at scale.
  • Industry-specific Cowork variants (legal, medical, financial).
  • Dramatic increase in task completion rates.

Long-term (3-5 years):

  1. AI agents as a standard operating environment. “Coworker” becoming literal AI colleagues, not tools.
  2. Fundamental reshaping of knowledge work.

The development velocity of AI is not going to slow down

Implementation Guide: Getting Started with Claude Cowork

If you are a Claude Max subscriber, here is what you should do:

Step 1: Download Claude Desktop (macOS)

  • Get it at claude.com/download.

Step 2: Turn on Cowork

  • Click on “Cowork” in the sidebar and choose a folder.

Step 3: Keep it Simple

Run simple examples at first:

  1. Organize the downloads folder.
  2. Extract expenses from receipts and list them in spread sheet.
  3. Turn notes into a document draft.

Step 4: Gradually Increase

  • Integrate Chrome for web, related tasks, connect to pre-existing tools, and try out complicated workflows.

Best Practices:

  1. Test with unimportant data (test folder).
  2. Look at what Claude is doing at the start.
  3. Give good, detailed directions.
  4. Give background and limitations.

Conclusion

Claude Cowork wasn’t formulated in 10 days solely due to Anthropic’s geniality (although they certainly are). It was the creation of 10 days because users had already validated the demand by pushing Claude Code to perform non-coding tasks.

The takeaway isn’t “AI accelerates development,” although it does. The takeaway is, “Make what users are already trying to create on their own, then simply get out of their way.”

Conventional product development allocates numerous years to create features that users might desire. On the contrary, AI-driven development is capable of releasing verified features within a few days.

This is not only about productivity tools. It’s about a big change in the manner in which software is developed:

  1. Planning to observe.
  2. From guessing to validating features to remove friction.
  3. From 10 years to 10 days.

The winning businesses thus won’t be the ones with the best plans. They will be the ones who see what users really do and deliver it quicker than everyone thought possible.

Claude Cowork demonstrates that it is feasible. The issue is: who is developing the next tool that users are already demanding?

Getting Started with Orbilon Technologies

At Orbilon Technologies, we guide enterprises to create AI-powered products in the right way, first by verifying the demand and then shipping rapidly with the help of AI and accelerated development.

Our Services:

  1. AI product strategy and validation.
  2. Rapid prototyping with AI development tools.
  3. Claude Cowork integration and automation.
  4. AI agent implementation for businesses.
  5. User behavior analysis and product discovery.
  6. AI-powered development acceleration.

We use the same method the Anthropic team applied with Claude Cowork: watch users, validate demand, build with AI, and iterate based on real usage. With our approach, we have enabled companies to deliver products 10 to 30 times faster.

If you want to create your product in days rather than years, come and visit us at orbilontech.com, or you can send an email to support@orbilontech.com for a discussion on how AI-powered development can shorten your product timeline.

Want to Hire Us?

Are you ready to turn your ideas into a reality? Hire Orbilon Technologies today and start working right away with qualified resources. We will take care of everything from design, development, security, quality assurance and deployment. We are just a click away.

Work with Us
Quick Connect