I got a private lesson on Claude Cowork & Claude Code

Understanding Claude Co-work and Its Potential

Introduction to Claude Co-work

• Claude Cowork is introduced as a tool that can help users outperform 99% of people by simplifying the use of Claude Code, which has gained popularity but is often seen as technical and challenging for beginners.

• The product aims to provide an easy-to-use interface (UI) for Claude Code, making it accessible for everyone, including those with little technical knowledge.

Guest Introduction: Boris

• Boris, the creator of both Claude Code and Claude Co-work, discusses his excitement about sharing best practices for using the new product effectively.

• He anticipates that listeners will gain insights into various use cases for Co-work and encourages them to share their experiences on social media.

Evolution of User Interaction with AI

• Boris reflects on how initial expectations for Claude Code were focused on coding, but users have found diverse applications beyond its intended purpose.

• He expresses curiosity about how users will interact with Co-work in unexpected ways, similar to past technological innovations where creators did not foresee all potential uses.

The Nature of Innovation

• The discussion highlights the unpredictable nature of app development; just as early apps led to major platforms like Uber and TikTok, current developments in AI may yield unforeseen applications.

• Both hosts agree on the importance of sharing experiences during this exploratory phase in AI technology.

Overview of the Desktop App Features

• Boris provides a walkthrough of the cloud desktop app's features, noting that Co-work is currently available only for Mac OS with Windows support coming soon.

• The app includes several tabs: chat (default), co-work (new feature), and code (Claude Code), emphasizing that Co-work operates using the same underlying technology as Claude Code.

Understanding Agentic AI

• Boris explains "agentic" AI—highlighting its ability to take actions beyond simple text interactions—and emphasizes its significance in enhancing user experience through practical tool usage.

• He notes that many existing products mislabel themselves as agentic when they do not fulfill this capability; thus, true agentic functionality is crucial for effective interaction.

Practical Applications Demonstration

• As part of demonstrating how to utilize Co-work effectively, Boris prepares to showcase specific functionalities within the application.

Understanding Co-Work and AI Integration

Accessing Files and Folder Permissions

• Users must opt-in to allow co-work access to specific folders on their desktop, as it cannot see anything by default.

• This represents a significant mindset shift; co-work operates similarly to an operating system, managing files that users permit it to access.

Capabilities of Co-Work

• Beyond file management, co-work can generate files such as presentations and interact with various tools via MCP (Multi-channel Communication Protocol).

• Initial steps for users include mounting a folder for co-work access and experimenting with its features for organizing files.

Interaction and Clarification

• The AI employs "reverse elicitation," asking for user clarification when uncertain about actions instead of making assumptions.

• After renaming receipts based on user input, the model demonstrates improved organization capabilities.

Safety Measures in AI Operations

• Co-work can take control of the user's computer with permission, emphasizing safety measures to prevent accidental data loss or deletion.

• Significant efforts have been made in AI safety at Anthropic, focusing on alignment and mechanistic interpretability to ensure safe operations.

Advanced Features and User Experience

• A virtual machine runs under the hood of co-work to safeguard actions taken by the model from affecting the broader system.

• New features like deletion protection prompt users before any accidental deletions occur, enhancing overall safety during interactions.

Real-Time Demonstration of Functionality

• During a demonstration, co-work successfully renamed receipts and created a spreadsheet based on user requests.

• The ability to treat co-work as a teammate allows users flexibility in task execution; however, this also presents challenges due to its extensive capabilities.

Browser Interaction

• Co-work opens browsers autonomously after receiving permissions from users during demonstrations, showcasing its integration into everyday tasks.

Exploring the Capabilities of AI in Everyday Tasks

Initial Experience with AI Ordering

• The speaker recounts an experience where an AI was tasked with ordering a pizza, specifically a pineapple pizza. The process was tedious and took about an hour to complete.

• Following this initial experience, improvements were made to enhance the model's ability to interact with computers, including spreadsheets and web browsers.

Enhancements in Spreadsheet Interaction

• The speaker emphasizes the importance of formatting when sending spreadsheets, noting that while the AI can format data, it still makes mistakes that need correction.

• There is excitement about how users will creatively utilize these tools in unexpected ways, similar to how smartphones evolved into platforms for apps like Uber.

General Purpose Applications of AI

• The discussion highlights the general-purpose nature of AI tools, comparing their potential impact on productivity to that of early smartphones.

• A focus on auditing company tasks arises; understanding how teams interact with files and use the internet can reveal opportunities for increased productivity.

Parallel Task Management

• The speaker mentions running multiple tasks simultaneously as a way to enhance efficiency and reduce time spent on tedious activities.

• There's a reflection on how engineers have been early adopters of such technology, automating repetitive tasks which allows them to focus on more enjoyable work.

Emailing Capabilities Demonstration

• As part of a demonstration, the speaker prepares to email a spreadsheet using Gmail. This showcases the model's ability to pull contacts and fill out information accurately.

• Anticipating criticism regarding speed compared to manual processes, the speaker acknowledges that while humans may be faster initially, parallel task execution by AI ultimately saves time.

Future Implications and User Adaptation

• The conversation concludes with thoughts on user adaptation; as models improve over time in speed and accuracy, they will become integral tools for various professional tasks.

Exploring the Future of Cloud Code and Co-Work

The Shift Towards Generalist Workflows

• The speaker discusses a shift in workflow dynamics, emphasizing a move towards generalist roles rather than deep specialization. This reflects a trend where individuals manage multiple tasks while ensuring their teams are aligned and informed.

Enhancements in Research Capabilities

• A new chat feature has been introduced that enhances research capabilities by sourcing information from various platforms, not just predefined websites. This includes checking notes and guidelines related to specific topics.

Streamlining Team Communication

• The speaker describes using automation tools to streamline team communication, particularly for tracking project statuses without needing constant follow-ups with team members. This allows for more efficient use of time.

Accessibility of Cloud Code

• There is a discussion about how cloud code, initially designed for terminal use, has evolved to be accessible through various platforms like mobile apps and Slack. This evolution makes it more approachable for non-engineers.

Surprising Adoption Among Non-Engineers

• It’s noted that many non-engineers at Enthropic actively use cloud code weekly, which was unexpected. Teams such as sales and design have integrated this tool into their daily workflows, indicating its broad appeal beyond technical users.

User-Friendly Interfaces Over Terminal Use

• The preference among users leans towards user-friendly interfaces rather than traditional terminal commands. Users appreciate having intuitive UIs that automate complex processes without requiring extensive technical knowledge.

Automation Aspirations with Skills Integration

• There is an aspiration to automate repetitive tasks through skills integration within the platform. Skills are seen as repeatable actions that can enhance productivity when properly utilized within co-work environments.

Customization vs Simplicity in Tools

• While engineers often prefer customizable tools due to diverse working styles and tech stacks, the initial approach for co-work aims at simplicity to cater to broader audiences. Users are encouraged to start with basic setups before customizing extensively.

This structured overview captures key insights from the transcript while providing timestamps for easy reference back to specific discussions or points made during the conversation.

Discussion on Co-Work and Coding Evolution

Initial Thoughts on Writing Skills

• The speaker suggests starting with simple tasks rather than focusing on writing skills, emphasizing the importance of understanding what is useful in early stages.

• They compare the current state of co-work to quad code from a year ago, noting that while both were released early and had bugs, co-work is already more functional and beneficial.

Growth and Future Predictions

• Reflecting on January 2026, the speaker speculates about how co-work will evolve by January 2027, acknowledging the rapid changes in technology.

• They express difficulty in making long-term predictions due to the exponential growth of models, indicating a preference for short-term planning.

Exponential Growth in Coding

• A prediction made last year suggested that people would stop writing code entirely; the speaker shares their experience where quad code has written all their code recently.

• They highlight that predicting such advancements requires an understanding of exponential growth rather than linear thinking.

Automation of Tedious Tasks

• The speaker believes automation will increasingly handle tedious tasks like data management, allowing individuals to focus on more enjoyable work.

• This shift could lead to higher productivity as users can delegate repetitive tasks to AI models.

Speculation on Co-Work's Role

• The speaker describes co-work as a "gateway drug" to coding tools, suggesting it will inspire specific use cases tailored to various industries.

• They foresee a future where digital teammates are integrated into workflows alongside human skills.

Technology's Rapid Advancement

• The discussion touches upon how technological evolution differs from past waves (like telephones or computers), highlighting its unprecedented speed and interconnectedness.

• The speaker emphasizes that AI's existence relies heavily on previous technologies' infrastructure and capabilities.

Viral Post Discussion

• Transitioning topics, they mention a viral post about Cloud Code that garnered significant attention (99,000 bookmarks), showcasing their learning journey with social media.

Cloud Code Setup and Workflow

Overview of Cloud Code Usage

• The speaker shares their relatively standard setup for using Cloud Code, emphasizing that it works effectively out of the box without much customization.

• There is no single correct way to utilize Cloud Code; users can customize and adapt it according to their preferences.

Task Management in Parallel

• The speaker discusses the importance of managing multiple tasks simultaneously rather than focusing deeply on one task at a time.

• They typically work in a terminal or mobile app, starting tasks in different tabs and switching between them as needed.

Planning and Execution

• Once a plan is established for coding tasks, the speaker quickly moves to auto-accept edits, noting improvements with Opus 4.5's ability to execute plans effectively.

• The excitement around Opus 4.5 stems from its enhanced capabilities in both coding and planning compared to previous models.

Development Methodology

• A quote highlights the relationship between planning and coding: "Once the plan is good, the code is good," indicating that effective planning leads to successful execution.

• The speaker reflects on spec-driven development, suggesting that any form of specification—like a simple text file—can suffice for effective development.

Multi-platform Coding Approach

• They mention running multiple cloud sessions in parallel while coding locally, utilizing both web interfaces and mobile apps for flexibility.

• Surprisingly, half of their coding now occurs on mobile devices—a shift they did not anticipate a year ago.

Recommendations for Effective Use

• Users are encouraged to leverage iOS and web applications simultaneously for optimal results when working with Cloud Code.

• Emphasizing environment setup's importance, they recommend investing time into maintaining an up-to-date CloudMD document shared among team members.

Insights on Model Efficiency

• The speaker praises Opus 4.5 as the best coding model available due to its efficiency; despite being larger and slower than smaller models, it ultimately saves tokens and costs less due to its intelligence.

Team Collaboration Practices

• Their team collaborates by sharing a single CloudMD repository checked into Git, ensuring continuous updates based on observed errors or inefficiencies during development.

Understanding CloudMD and Its Usage

Overview of CloudMD

• CloudMD is simply a text file without any special formatting requirements, allowing users to input information freely.

• During code reviews, the speaker tags colleagues' pull requests (PRs) to update the CloudMD as part of the process, utilizing GitHub actions for efficiency.

Compound Engineering Explained

• The term "compound engineering" is clarified; it’s important to note that it should be referred to correctly as "compound engineering," not "compounding engineering."

• The speaker describes how to install the Claude app in a GitHub repository using a specific command, enabling easy updates and changes through mentions.

Practical Applications of CloudMD

• Frequent use of tagging Claude helps keep the knowledge base updated and minimizes repetitive comments during code reviews.

• A previous method involved maintaining a spreadsheet of recurring issues from code reviews, which led to creating lint rules for automation—now replaced by tagging Claude for similar outcomes.

Maximizing Efficiency with Quad Code

Planning Mode Utilization

• Most sessions begin in planning mode; this allows back-and-forth interaction with Claude until an acceptable plan is established before switching modes for execution.

• Emphasis on planning being underutilized despite its importance; many users are encouraged to adopt this feature more frequently.

Performance Improvement Tips

• Recommendations for enhancing performance include using Opus 4.5 with thinking capabilities and ensuring good quality inputs (quad).

• Allowing Claude to verify its output significantly improves results; practical examples include using Chrome extensions for testing outputs effectively.

The Importance of Verification in Engineering

Enhancing Output Quality

• The analogy of a painter working blindfolded illustrates how crucial it is for engineers (or AI models like Claude) to see their outputs in order to improve quality.

• As AI models advance, providing them with verification methods will enhance their initial output quality significantly.

Encouragement for Exploration

• Users are encouraged to experiment with different workflows when using these tools, likening it to interactive storytelling where one can choose their path.

Insights on Productivity and Collaboration in Startups

The Importance of Flexibility in Work Methods

• The discussion emphasizes a free-form approach to productivity, suggesting that there is no singular method for achieving success. Individuals are encouraged to explore what works best for them.

• Boris shares insights on maximizing the use of co-work and Claude code, highlighting its relevance to many listeners who are eager to enhance their productivity.

Audience Engagement and Relevance

• The podcast audience consists of millions seeking ways to increase productivity and build businesses around innovative ideas. This reflects a broader trend in startup culture focused on efficiency.

• Acknowledgment of Boris's limited but impactful social media presence suggests that his insights are valuable when shared.

Cultural Nuances in Language

• A light-hearted moment arises as the speaker mentions calling Claude "Clo," revealing cultural differences in language usage, particularly from French Canada.

• Boris responds affirmatively, indicating that he believes "Clo" sounds better than other pronunciations, showcasing personal preferences influenced by cultural context.