How to Make Claude Code Better Every Time You Use It (50 Min Tutorial) | Kieran Klaassen

AI Learning and Compound Engineering

Introduction to AI Learning

• AI has the capability to learn from user interactions, improving its performance over time by understanding preferences and mistakes.

• The process allows for immediate testing and validation of changes, akin to a QA team that can fix issues on-the-fly.

Guest Introduction: Kieran

• Kieran, the CTO of Kora, is introduced as an expert in cloud code and compound engineering.

• The discussion highlights the growing popularity of cloud code within social media conversations.

Overview of Compound Engineering Philosophy

• Kieran introduces "compound engineering," a philosophy he developed while building Kora, focusing on best practices learned from AI usage.

• The philosophy emphasizes that AI can learn effectively when users invest time in training it with feedback.

Four Steps of Compound Engineering

• The four steps include:

• Planning: Establishing what needs to be done.

• Working: Executing tasks based on the plan.

• Assessing: Reviewing completed work for accuracy.

• Codifying Learnings: Capturing insights for future reference.

Practical Application of Compound Engineering

• Emphasizes the importance of capturing learnings from each cycle to improve subsequent planning processes.

• A plugin exists that supports this philosophy, allowing users to implement these steps in their projects seamlessly.

Building Agent Native Features

Project Goals with Kora

• Kora aims to create an email assistant capable of performing all user actions within the app settings autonomously.

Workflow Planning Process

• The planning phase involves defining workflows where the assistant mimics user capabilities in changing settings through voice commands instead of typing.

• Users are encouraged to follow along with practical examples using specific commands related to compound engineering.

Granola: The AI Meeting Notes App

Introduction to Granola

• The episode is sponsored by Granola, an AI meeting notes app designed to streamline note-taking during meetings.

• Granola automatically captures notes and allows users to add their own insights, making post-meeting organization easier.

• Users can extract clear takeaways and next steps in a preferred format using Granola's features.

Sharing and Efficiency

• Notes can be shared directly via Slack or set up for automatic sharing, enhancing team collaboration.

• The speaker emphasizes that among various AI tools, Granola saves the most time.

Understanding Code Development with AI

Behind the Scenes of Code Development

• Discussion shifts to how AI assists in code development through methods like spec-driven development and task management.

• The AI analyzes existing code to identify patterns and best practices from online resources.

Framework Compatibility

• It checks compatibility with specific frameworks and libraries, ensuring correct version usage to avoid mismatches.

Token Usage and Cost Considerations

Token Management

• The conversation highlights that certain agents run as sub-agents with separate contexts, consuming significant tokens (e.g., 56,000 tokens).

• To utilize these features effectively, users need a max plan due to high token consumption but find it valuable for professional coding tasks.

Investment vs. Cost

• Acknowledgment of the cost-benefit analysis where investing in advanced tools is justified by potential productivity gains for developers.

Future of Development Tools

Embracing Change

• Encouragement for developers to experiment with new models rather than being stuck in traditional methods; this shift represents a significant mindset change.

Long-term Goals

• The ultimate aim is automation where minimal input leads to complete outputs without needing detailed understanding of every step involved.

Best Practices in Coding

Customization and Learning

• Developers are encouraged to understand underlying processes by customizing plugins or flows based on their needs.

Researching Best Practices

• Utilizing research capabilities within tools enhances coding efficiency beyond basic functionalities available in standard modes.

Collaboration Insights

Team Dynamics

• Discussion about team members' contributions highlights collaborative efforts leading to rapid advancements (e.g., shipping multiple PR requests).

Personal Productivity

• Reflection on personal productivity shows reliance on AI tools has increased significantly over time.

Agent Native Architecture and Skills Overview

Introduction to Warp and Agent Native Architecture

• The speaker expresses a preference for using Warp, highlighting its ability to fix issues with Git commands easily.

• Discussion on the importance of agent native skills, which distill specific information into manageable formats for easier access.

Understanding Skills in Agent Native Architecture

• The agent native architecture document is extensive, providing detailed guidance on app development but can overwhelm context if loaded entirely.

• Skills are designed to be context-aware; they pull relevant information when needed without overloading the system.

• Users can create custom skills tailored to their needs (e.g., iOS development), enhancing the functionality of the compound engineering plugin.

Utilizing Spec Flow Analyzer

• The spec flow analyzer helps ensure that all necessary components are considered during planning by simulating user personas.

• It identifies potential gaps in planning and assists in writing comprehensive plans based on user feedback.

Plan Review Process

• After completing a plan, users have options such as opening it in an editor or deepening the plan for thorough analysis.

• Deepening the plan allows for extensive review using available skills, optimizing token usage effectively.

Enhancing User Experience with Tools

• Users can share plans with teams for feedback and utilize aesthetically pleasing apps like Typora for better readability of markdown documents.

• The speaker emphasizes the importance of visual appeal in documentation to enhance engagement and understanding.

Final Thoughts on Engineering Practices

• Acknowledges that while some aspects may be complex, non-engineers can still navigate processes effectively without needing deep technical knowledge.

Discussion on Tool Consolidation and Compound Engineering

Exploring Tool Efficiency

• The speaker discusses the challenge of managing multiple engineering tools, suggesting a need for consolidation to streamline processes.

• They propose creating a single tool capable of handling multiple settings to reduce complexity and improve efficiency.

Importance of Early Implementation

• Emphasizes the significance of addressing potential issues early in the development process to avoid complications later on.

• The speaker mentions using a "compound flow" to document insights, which helps prevent repeated mistakes in future planning.

Documentation Practices

• A dedicated documentation directory is mentioned, where learnings are organized for easy access during planning phases.

• The option to update documentation files (cloth MD file) is highlighted as an efficient way to capture knowledge without needing extensive plugins.

Trusting AI in Development

• The conversation shifts towards trusting AI systems for code generation and documentation updates, contrasting traditional engineers' skepticism with newer approaches that embrace automation.

• The speaker notes a generational shift where some engineers prefer not to scrutinize every line of code generated by AI, relying instead on established systems.

Balancing Oversight and Innovation

• While acknowledging the importance of oversight in critical projects (like Kora), they advocate for a more relaxed approach when experimenting or working on less critical tasks.

• The concept of "compound engineering" is introduced as a method for capturing learning experiences so that developers do not have to repeatedly check their code.

Final Steps in Workflow Execution

• After compounding knowledge, the next step involves executing workflows confidently with updated plans based on consolidated tools.

• The discussion concludes with an emphasis on maintaining clear context while working through iterations efficiently.

Understanding the Integration of Tools in Development

Tool Selection and Integration

• The speaker discusses using Markdown for planning but emphasizes flexibility, allowing users to choose tools like GitHub or Linear based on their preferences.

• A question arises about the difference between simply instructing to "go build it" versus a structured approach, highlighting the importance of context in development tasks.

Planning and Contextual Awareness

• The initial phase involves identifying missing elements in a project plan, ensuring that all necessary components are addressed before proceeding.

• The discussion acknowledges that not all questions can be answered definitively, indicating the complexity of certain development scenarios.

Thought Process and Management Skills

• Emphasizes a "thinking mode" where potential issues are anticipated, promoting proactive problem-solving rather than reactive measures.

• The interaction resembles managing an engineering team, requiring tech lead skills to guide agents through complex tasks effectively.

Future of Development Automation

• Anticipation of a future where minimal input leads to substantial results is discussed; likening this process to onboarding new team members.

• Once trained, automated systems could operate independently, streamlining workflows significantly.

Testing with Opus 4.5: Enhancements and Capabilities

Advancements in Testing Features

• Opus 4.5 introduces improved testing capabilities using Playwright, enhancing automation in testing processes compared to previous models.

• The speaker expresses satisfaction with Playwright's performance in automating browser interactions for testing purposes.

Practical Application of Playwright

• Playwright allows control over Chrome windows via code integration; it supports recording sessions and generating visual documentation automatically.

• Installation is simplified to one command line entry for ease of use; once set up, it automates browser interactions seamlessly during tests.

Continuous Improvement and Feedback Loop

• Discussion on how running tests generates plans detailing new features introduced within the codebase while maintaining clarity on what needs testing.

• Highlights Opus's ability to follow directions effectively without needing excessive oversight or intervention from developers.

Testing Features in Software Development

The Importance of Browser Testing

• The discussion highlights the significance of browser testing as a crucial component in software development, serving as the ultimate test to determine if features function correctly.

• Emphasizes that this testing approach requires no extensive coding or overhead in continuous integration (CI), effectively acting like a quality assurance (QA) team.

• The ability to iterate quickly is noted, allowing immediate fixes and validations when issues arise during testing.

Capabilities of Automated Testing Tools

• Automated tools can interact with web elements, execute JavaScript, and access console logs, providing comprehensive control over the browser environment.

• A practical example is given where integration tests are simplified by logging into Gmail through an automated tool to verify email signature features.

Assessing Security and Code Quality

• After initial testing, it’s essential to assess for security risks and code simplicity; this stage involves reviewing from multiple perspectives.

• An agent-based review system is introduced that incorporates various viewpoints including security and architectural considerations.

Synthesizing Feedback for Improvement

• The feedback gathered from reviews creates a prioritized list of improvements without immediately implementing changes, allowing for thoughtful consideration.

• It’s important to maintain focus on the original plan while being open to insights that could enhance the project.

Managing To-Dos and Action Items

• A structured approach is taken towards managing tasks with different priority levels (P1, P2, P3), ensuring critical issues are addressed first.

• A command called "triage" helps streamline decision-making by guiding users through findings conversationally before executing resolutions automatically.

Workflow Automation and Command Efficiency

Overview of Workflow Automation

• The speaker discusses the flexibility of using a system for workflow automation, emphasizing that users can choose to implement only parts of it, such as the planning phase, while still enjoying manual coding.

• A demonstration is provided on resolving tasks in parallel through a dependency graph, showcasing how sub-agents manage to-dos efficiently and create pull requests quickly.

Simplifying Commands with LFG

• Introduction of the "LFG" command which encapsulates multiple actions into one command, streamlining processes significantly.

• The LFG command not only executes previous steps but also generates marketing materials like videos and changelogs automatically.

Agent Orchestration Insights

• Discussion on the importance of having a robust system for agent orchestration, likening it to combat engineering for effective task management.

• The speaker highlights how setting up these systems simplifies workflows and enhances productivity.

Managing Permissions Effectively

• A solution is presented for managing permissions without constant approval prompts by using an alias called "dangerously skip permissions."

• Caution is advised when using this mode; users should ensure their environment is secure to prevent accidental data loss or security breaches.

Enhancing User Experience with Cloud Integration

• Users can add permissions permanently through settings files, making future interactions smoother after an initial adjustment period.

• Assurance that the system (Opus 4.5) operates intelligently without performing harmful actions like deleting important files unexpectedly.

Utilizing Mobile Applications for Continuity

• The speaker explains how commands can be executed from mobile devices via cloud integration, allowing seamless transitions between desktop and mobile environments.

• Demonstration of continuing work on pull requests through a mobile app despite some bugs present in its interface.

Understanding Workflow Automation with Slash Commands and Sub Agents

Overview of Command Usage

• The discussion begins with the ability to open a command line interface (CLI) from a specific location, allowing users to continue conversations seamlessly between web interfaces and CLI.

• The complexity of slash commands, sub agents, hooks, and skills is acknowledged as overwhelming for learners trying to grasp these concepts.

Functionality of Slash Commands

• Slash commands are used to trigger workflows that can call sub agents. This structure allows for organized task management within the system.

• The speaker emphasizes that slash commands serve as direct instructions for actions like creating pull requests or committing changes.

Role of Sub Agents

• Sub agents are designed for specialized tasks where full context isn't necessary; they provide quick results without needing extensive background information.

• They are particularly useful for parallel processing, enabling multiple tasks to be executed simultaneously.

Triggering Mechanisms

• While it’s possible to manually trigger sub agents, the preferred method is through slash commands which encapsulate business logic effectively.

• The speaker clarifies that they do not typically add sub agents manually but rely on structured command triggers instead.

Skills Integration

• Skills allow additional functionalities such as image generation via APIs. They enhance capabilities by injecting knowledge dynamically into the workflow.

• Specificity in invoking skills is crucial; clear instructions lead to better performance when using cloud applications.

Video Generation Process

• A marketing video is being generated using Playwright, which records screen activity and uploads it automatically as part of a pull request review process.

• This feature enhances code reviews by providing visual documentation alongside textual changes, making collaboration more efficient.

By structuring the notes this way, key insights from the transcript are highlighted while maintaining clarity and ease of navigation through timestamps.

Compound Engineering GitHub Plugin Overview

Accessing the Compound Engineering Plugin

• The compound engineering GitHub plugin provides comprehensive access to various features and commands, allowing users to explore its functionalities.

• Users can easily copy specific commands, such as those related to feature videos, directly into their local directories for convenience.

• The speaker expresses excitement about the automation capabilities of the plugin, noting that it reduces manual tasks significantly.

Testing Features with Playwright

• The integration of Playwright within the plugin is highlighted as a remarkable feature, enabling users to start testing efficiently and create video content seamlessly.

• There is an emphasis on how automation through this tool enhances productivity by taking over repetitive tasks that were previously done manually.

Additional Resources

• A link to the compound engineering plugin will be provided in the description for easy access.