What I’m Learning

This is where I keep track of what I’m learning these days. Sources become structured JSON first, then this converter turns accepted notes into Quartz Markdown so backlinks, graph view, tags, search, explorer, callouts, and wikilinks stay native to Quartz.

Pipeline

source -> LLM JSON -> Markdown converter -> Quartz content -> Quartz build -> /learning

AI

• Agent-First Software Development - Explores the paradigm shift in software development where AI agents, like Codex, are the primary code generators, with humans focusing on system design, intent specification, and feedback loops.
• Codex - Codex is an AI model developed by OpenAI, based on GPT, that translates natural language into code. It was used extensively in an experiment to generate an entire software product.
• Large Language Models - An overview of Large Language Models (LLMs) as foundational technology for advanced AI capabilities like code generation and natural language understanding.
• Managed Agents - An overview of Anthropic’s Managed Agents system, which decouples the LLM ‘brain’ from its ‘hands’ (sandboxes and tools) and ‘session’ log to enhance scalability, reliability, and security.

DevOps Tools

• Using Terraform to Bootstrap GitHub CI Processes - Leveraging Terraform to automate the initial setup of CI/CD pipelines, including GitHub Actions environments, repository variables, and Azure permissions.

Observability

• Service Topology: Real-Time Service Dependency Mapping - Netflix’s approach to building a real-time, multi-layered service topology map to visualize dependencies, troubleshoot faster, and understand blast radius in large microservice architectures.
• Understanding SLIs, SLOs, and SLAs for Service Reliability - Clarifying the distinct roles and practical applications of Service Level Indicators (SLIs), Service Level Objectives (SLOs), and Service Level Agreements (SLAs) in managing and improving service reliability.

Sources

• Harness engineering: leveraging Codex in an agent-first world - OpenAI experimented with building a software product entirely with Codex agents, writing 1 million lines of code with 0 manually-written lines, drastically increasing engineering velocity and redefining the engineer’s role to system design and agent supervision.
• Scaling Managed Agents: Decoupling the brain from the hands - Anthropic’s Managed Agents system decouples the LLM ‘brain’ from its ‘hands’ (sandboxes and tools) and ‘session’ log to improve scalability, reliability, and security, drawing parallels to operating system abstractions.

System Design

• Abstraction - The principle of abstraction in system design, where complex systems are simplified by hiding underlying details behind interfaces, enabling flexibility and longevity.
• Decoupling - The system design principle of decoupling, which involves reducing dependencies between components to improve modularity, flexibility, scalability, and fault tolerance.
• Enforcing Architectural Invariants - The practice of mechanically enforcing architectural rules and boundaries within a codebase, especially crucial for agent-generated software.
• Repository Knowledge as System of Record - Treating the code repository, including structured documentation, as the definitive source of truth for system knowledge, crucial for AI agent operation.