Person/Michael Yong/Blog/25/07/Claude Code Unleashed

• Claude Code Unleashed - Anthropic/AI/App/Claude Code

  • My Notes
    • 2025-07-18 Fri
      • Person/Michael Yong really Gets It when it comes to the future of coding.
      • These are his lessons learned building Background Agents while trying to make Claude/Code/Tool/Claude Code Terragon

      • Cost Savings

        • I didn't realize that one could get savings from Anthropic/Product/Claude/Max/20x vs Anthropic/Product/Claude individually billed.
          • 
        • Key Insight

          • I highly recommend getting a claude max subscription. Not having to worry about rate limits, per-use cost, whether you should use sonnet or opus is truly a game changer. With Max, I stopped thinking in terms of "is this a task that's good for claude code? Is this a task for sonnet or opus?" Instead, it became "How can I offload more work to claude code?" That mental shift alone was worth the price. - #Quote from Person/Michael Yong
            

            • reminds me of IndyDevDan's Compute Advantage equation and ideas
              • Person/IndyDevDan/Compute Advantage

                • Compute Advantage Insight

                  • Compute Advantage = The higher your compute advantage, the more value you can produce as an engineer using compute as leverage.
                  • Increasing compute advantage examples, in order
                    • using 1 prompt to make an application
                    • using multiple prompts in a row to make an application
                    • using multiple prompts in a loop to make an application -
                    • using prompts with tool calls in a loop to make an application - Agentic System
                    • combining a reasoning model with powerful execution models in an agentic system that makes tool calls in a loop to make an application
              • Person/IndyDevDan/Compute Advantage/Equation

                • Compute Advantage Equation and Calculator - via Person/IndyDevDan/YouTube/25/04/One framework for evaluating AI Coding

                  • Compute Advantage = ( Compute Scaling × Autonomy ) ÷ ( Time + Effort + Monetary Cost )
                  • Definitions
                    • Compute Scaling: Raw intelligence and capability of your model
                    • Autonomy: How much/well your AI tools can operate without you
                    • Time: Calendar time from start to completion
                    • Effort: Mental energy you spend on the task
                    • Monetary Cost: Direct financial expense of tools and services
                  • Insight
                    • The higher your compute scaling and autonomy, and the lower your time, effort, and monetary cost, the greater your compute advantage as an engineer.
      • Anthropic/AI/App/Claude Code and Git Worktrees
        • he says it's a pain, and I imagine it is

          • You can't help but feel like you're constantly juggling:

            • What was I doing?

            • Which branch am I on?

            • Which tab was this again?

            • What was the context of this change?

            • Which IDE window corresponds to this worktree?

        • This is the inspiration for the thing he's building, Claude/Code/Tool/Claude Code Terragon
          • #Goals

            • In the past week, I've averaged ~30 tasks running on terragon a day. Locally, I'm often testing and finishing up a task that was started by a background agent.
              

          • Remind/s Me of
            • Latent Space/Pod/25/06/Solomon Hykes with Dagger to Contain Agent Chaos/Agentic Impetus

              • Person/Solomon Hykes, the founder of Docker and developer of Dagger explains why Container/iz/ation is key to AI Coding

                • • AI/Summary of relevant parts at 11:25

                    • Hykes explained that while Docker's foundational container technology revolutionized environment portability, the tools built around it—like Dockerfiles and Compose—were designed as temporary solutions a decade ago and "frozen in time," inadequately serving today's needs for agent-native developer workflows.
                    • He emphasized that as AI coding agents become more central, environment reproducibility is critical so agents can operate in isolated, portable, and observable environments.
                    • Dagger aims to solve these problems by building an agent-native, modular workflow engine on existing open standards (containers, Git, OpenAI APIs) instead of retrofitting legacy Docker interfaces.
                    • Hykes noted that while Docker made containers ubiquitous, it isn't "agent native," and the next wave of developer tooling must focus on isolating agent environments to avoid vendor lock-in, enabling local and cloud execution, and providing reproducibility for both agents and humans across platforms and infrastructures
      • #Learned
        • AI/Coding/Idea/25/07/Tools for Context Switching
          • In time, I imagine that we will develop assistive technologies specifically to aid us with context switching in a concurrent workstream world!

          • Managing multiple agents requires a lot of context switching. This isn't normal! You're suddenly tracking 10+ concurrent work streams instead of 1-2. But you get to do a lot more things in a shorter amount of time. -- #Quote from Person/Michael Yong
            

      • AI/Coding/Problem/Task Decomposition

        • Using Claude/Code/Tool/Claude Code Terragon has fundamentally changed how I think about breaking down work. I'm creating smaller, more agent-friendly tasks. #Quote from Person/Michael Yong
          

          • Remind/s Me of
            • asdf