The Four Step Process to Loop Engineer ANYTHING (+ Why Prompt Engineering Isn't Dead)

Too Long; Didn't Watch — Summary

Loop engineering is an iterative framework of stacked prompts designed for self-improvement and goal-attainment, rather than a replacement for prompt engineering.

Main Takeaways

• Prompt Engineering is Not Dead: Loops are essentially prompts repeated with scaffolding; understanding how to prompt remains the foundational skill.
• The Four-Phase Framework: Every effective loop requires a Trigger, Execution (skills), Verification (success criteria), and State (memory/logging).
• Success Criteria is Paramount: Loops excel when goals are objective (e.g., code runtime); "fuzzy" goals (e.g., content quality) require more complex verification like human-in-the-loop or secondary LLM judges.
• State and Memory: For a loop to be truly "engineered," it must record its history to analyze what worked and what failed, ensuring each iteration is better than the last.
• The Automation Journey: Users should progress from manual tasks to skills, then to simple automation, before attempting complex loop engineering.

Detailed Summary

Introduction (0:00 - 0:50)

The current discourse suggesting prompt engineering is obsolete is inaccurate. Loops are simply tools—like a wrench compared to a screwdriver—and are built entirely upon prompts. Loop engineering is the process of setting up iterative tasks that repeat until a defined success criterion is met.

Defining the Four Phases (0:50 - 6:33)

• Trigger: The mechanism that starts the loop, such as a scheduled cron job, web hook, or routine.
• Execution: The phase where the AI performs the task, ideally using specific "skills" to ensure consistent outputs.
• Verification: The most critical phase involving success criteria. Objective criteria (like Python runtime) are ideal, while subjective criteria (like LinkedIn engagement) are "fuzzy" and harder to automate.
• State: The memory component. A loop must log its outputs and results to a database or document so subsequent runs can learn from previous successes and failures.
• Stop Criteria: Essential for cost management; loops need a defined end point, such as reaching a goal or hitting a maximum number of iterations.

Loop Engineering vs. Existing Tools (6:33 - 10:00)

• Auto-Research: While similar, tools like Karpathy’s auto-research require strictly objective, numerical success criteria.
• Claude Code /goal: This command is a micro-version of loop engineering but is limited to a single session. True loop engineering has an "infinite horizon," meaning it runs indefinitely across multiple sessions to provide continuous improvement.

The Four-Step Implementation Process (10:00 - 16:16)

1. Manual Process: Verify the AI can actually perform the task manually before trying to automate it.
2. Codify into a Skill: Turn the manual prompts into a reusable skill for specific, repeatable outcomes.
3. Automate the Skill: Use routines or schedules to run the skill without manual intervention.
4. Loop Engineering: Add success verification and state logging to allow the system to analyze its own performance and self-improve.

Handling Fuzzy Goals and Complexity (16:16 - 19:21)

Using a LinkedIn content creation example, the complexity of "fuzzy" loops is explored. Because there is a delay between posting and receiving engagement data (likes), the loop requires a secondary process to scrape data and update the state database. This highlights that while objective loops (like code optimization) are straightforward, subjective loops require careful architectural planning regarding how the AI judges its own work.

Conclusion (19:21 - 20:01)

Loop engineering is a powerful method for creating self-improving AI systems, but it relies heavily on the quality of the underlying prompts and the clarity of the success metrics.

Notable Quotes

"A loop at its core is still a prompt. It's just a prompt that we are repeating over and over again with some additional scaffolding."

"Just because we discovered what a wrench is yesterday doesn't mean we throw out the screwdriver."

"If you don't have a strong goal and you don't have clear success criteria, this is all pointless and you're just going to be spinning your wheels and burning tokens."