The 4-Step Agentic Engineering Workflow I Use to Ship Real Apps | Micky Shimeles
Watch now (47 mins) | Micky's 4-step agentic engineering workflow and the open-source skills that make AI agents prove their work
This article details Micky’s 4-step agentic engineering workflow designed to prevent the creation of low-quality code. The process involves isolating each task for agents, building based on a clear structure, proving functionality through validation loops, and shipping after agents review or escalate pull requests. This workflow ensures that all code shipped to users undergoes rigorous testing and review.
- Micky’s agentic engineering workflow consists of four steps: Isolate, Build, Prove, and Ship.
- Step 1 (Isolate) involves giving each agent its own copy of the code to enable parallel processing.
- Step 2 (Build) focuses on writing code with a service-layer architecture, clear inputs, and structured outputs.
- Step 3 (Prove) uses evidence-driven testing where agents screen-record their testing process.
- Step 4 (Ship) involves using an AI reviewer like Greptile to test and improve pull requests before a potential manual review.
- The article also briefly touches upon topics like AI skills becoming obsolete and comparisons between coding tools like Astra and Fable.
https://bender.layer3.press/articles/843cdfa1-cd91-4357-b737-05a574feee94
Write a comment