How I AI: My Weekly Codex Experiments — AI Summary

Summary

Traditional prompt engineering is evolving; top AI users now treat their local file systems as dynamic context windows, particularly with models like Codex. This enables complex, multi-document tasks and fosters a truly collaborative interaction, drastically improving efficiency for long-form work.

Key Topics

• Context Window Assembly: Context window assembly involves intelligently gathering and organizing data from a local file system into a dedicated working folder. For models like Codex, this means instructing the AI in natural language to find and copy specific files (e.g., documents, spreadsheets, code) that are relevant to the current task. This curated context window is crucial because it allows the AI to focus its processing power on pertinent information, enabling it to handle much larger and more complex tasks (like processing 50,000 words or intricate codebases) with high accuracy and efficiency, far beyond what a single-prompt context window typically allows.
• Collaborative Prompting: Collaborative prompting is a shift from one-way instruction to a dynamic, back-and-forth interaction. Instead of providing a rigid, fully structured prompt, the user first asks the AI to help define the scope and parameters of the task, often by posing meaningful questions about desired standards. This allows the AI to contribute to the strategic planning, ensuring a shared understanding of the objectives before diving into execution. This approach leverages the strengths of advanced LLMs, which are less prone to 'getting lost' in complex, multi-step tasks when given the opportunity to help shape the workflow collaboratively, leading to more robust and aligned outcomes.
• Agentic Execution: Agentic execution refers to an AI's capability to take a high-level task and break it down into sub-tasks, plan a sequence of actions, and then execute those actions with some degree of autonomy. This is contrasted with simple prompt-response interactions. When combined with collaborative prompting, the AI can first help shape the task and then 'go and get it done' by performing multiple steps, interacting with resources, and even self-correcting based on internal or external feedback (like an auto-review system). This empowers users to delegate more complex, multi-stage projects to AI.
• Multi-threading with AI: Multi-threading with AI involves leveraging the AI's capabilities to manage and progress several distinct but potentially related tasks or creative projects at the same time. This could mean simultaneously drafting different versions of a document, developing multiple lines of code, or incubating several distinct ideas within a local folder structure. This approach is facilitated by models that can efficiently use file systems and maintain task focus over extended periods. It dramatically boosts productivity by allowing users to parallelize work, reducing bottlenecks and accelerating the development and refinement of complex outputs.
• Auto-Review Systems: Auto-review systems are integral to advanced AI workflows, especially when delegating long-running or multi-threaded tasks to an AI agent. These systems act as 'guardrails,' evaluating the AI's progress and output against predefined standards or objectives. They are crucial for maintaining quality control, identifying potential deviations, and ensuring that the AI remains aligned with the user's intentions without constant human oversight. This allows users to confidently let AI models run on their machines, knowing that a system is in place to manage and validate the work.

Key Takeaways

• Experiment with Codex or similar LLMs for local file system integration to manage large context windows.
• Adopt a collaborative, iterative prompting approach by asking the AI to help define task parameters before execution.
• Explore AI auto-review systems and multi-threading capabilities for complex projects.