Vibe Check: Claude Cowork Is Claude Code for the Rest of Us
Timestamps are as accurate as they can be but may be slightly off. We encourage you to listen to the full context.
Podcast Summary
In this live vibe check, Dan Shipper and Kieran Klaassen explore Anthropic's newly released Claude Cowork—essentially "Claude Code for everyone, not just engineers." (01:00) Felix Rieseberg, a product engineer from Anthropic who helped build Cowork, joins the conversation to explain the design philosophy behind this new interface. (31:06) The episode demonstrates real-time testing of Cowork's capabilities, from competitor analysis to email drafting and calendar audits, showing how it enables async workflows instead of traditional turn-by-turn chat interactions.
- Core theme: The evolution from chat-based AI interactions to agent-native, async workflows for knowledge workers
Speakers
Dan Shipper
Dan is the co-founder and CEO of Every, a media company and app studio at the edge of AI. He's also the author of the Agent Native Architectures guide and has been building AI-powered applications, including recently creating a markdown editor called Proof over a weekend using Claude Code.
Kieran Klaassen
Kieran is a developer at Every who builds Quora, an AI email assistant. He's deeply experienced with Claude Code and focuses on creating plugins, skills, and VST development tools, bringing a technical perspective to AI product evaluation.
Felix Rieseberg
Felix is a product engineer at Anthropic who was instrumental in building Claude Cowork. Previously at Slack for many years, he brings experience in building collaborative tools and understanding user workflows in professional environments.
Key Takeaways
Embrace Async AI Workflows
Claude Cowork represents a fundamental shift from synchronous chat interactions to asynchronous task management. (02:54) Unlike traditional Claude where you send a prompt and wait for a response, Cowork allows you to queue multiple tasks and messages while the AI is working. Felix explained that this enables users to "hand off something and then let it do stuff for thirty minutes or an hour and then come back and review that." This async approach mirrors how developers already work with Claude Code, but now extends these capabilities to non-technical users for research, data analysis, and document creation.
Skills Are the Primary Hackable Surface
According to Felix, skills have become "the primary hackable surface" for customizing AI behavior. (44:48) Instead of building specific MCP tools or complex harnesses, users can write skills in markdown that describe how Claude should approach tasks. Kieran demonstrated this by creating a skill that combines Swiss design principles with 3D printing capabilities to generate STL files. Skills automatically load from Claude into Cowork, making them a powerful way to personalize and extend the platform's capabilities across different interfaces.
Agent-Native Architecture Principles
The conversation revealed core principles for building agent-native applications: parity (agents should be able to do whatever users can do through UI), granularity (tools should be lower-level than features), composability (combining tools in unexpected ways), and emerging capabilities (discovering new use cases through user experimentation). (50:13) Felix emphasized that pushing tools into more general, composable spaces allows applications to benefit more from improvements in model intelligence rather than constantly building new specific tools.
Computer Access Enables New Workflows
Cowork's ability to control Chrome and access local files opens up entirely new categories of tasks. (08:42) Dan demonstrated having Claude analyze PostHog analytics by browsing through the interface while logged in, eliminating the need for API integrations. Similarly, Kieran showed how connecting Chrome allows the AI to work with any web-based tool where you're already authenticated, dramatically expanding what's possible without complex setup or technical knowledge.
Rapid Iteration Through Early Release
Anthropic's decision to ship Cowork as a separate tab after just "a week and a half" of development demonstrates a new approach to product development. (36:45) Felix explained they wanted to "ship almost every single day some new features, some bug fixes" and work closely with users rather than develop in isolation. This approach of releasing experimental features in dedicated spaces allows for faster iteration while managing user expectations about polish and stability.
Statistics & Facts
- Claude Cowork was built in just a week and a half according to Felix from Anthropic (36:45), demonstrating the rapid pace of AI product development.
- The "Chat with Claude" button on Every's agent native guide received 4,000 clicks according to PostHog analytics that Claude analyzed (09:23), showing significant user engagement with AI integration features.
- Nearly 10,000 people joined the live stream to watch the Claude Cowork demonstration (60:41), indicating strong interest in new AI product releases.
