My Agentic Engineering Workflow (after 15,700 sessions)
Summary
AI summaries can be incomplete or wrong. Verify anything important against the original video.
David Ondrej shares his evolved agentic engineering workflow, showcasing how he leverages advanced AI models, custom agent orchestration, and automated loops to achieve a high rate of production. The video covers his practical setup, including model selection, code review strategies, and the philosophy of building self-improving software.
In this extensive guide, David Ondrej presents a detailed breakdown of his current agentic engineering workflow, an evolution from his previous practices. He emphasizes that the 'harness' (the platform running the model) is becoming less important than the capability of the AI model itself, and he identifies models like Opus 5.5 and Astra as the current benchmarks for performance. He walks the audience through his preferred subscription models, explaining how he maximizes value across different agentic tools while highlighting the unique advantages of his open-source project, Cloudroom, which he designed to support multiple agent harnesses.
The video then shifts to the philosophy of being an 'agentic engineer,' arguing that modern engineering requires broader technical breadth rather than deep specialization in syntax. He explains that knowing how technologies work at a high level allows engineers to effectively guide agents to solve complex problems. A core part of his approach involves 'agent loops'—automated workflows that handle repetitive tasks like bug reporting, security sweeps, and funnel analysis. He explains how he identifies these tasks by scanning his own activity and converting the most repetitive segments into autonomous, self-improving agentic workflows.
Finally, Ondrej delves into the infrastructure choices, contrasting cloud-based vs. local inference. He advocates for cloud-based agents for the vast majority of tasks due to their ability to run 24/7 without being limited by local machine availability. He acknowledges the importance of self-hosted inference for sensitive projects, detailing his own hardware setup that allows for private execution of proprietary code or legal documents. The video concludes by encouraging viewers to build their own systems, not by blindly copying others, but by automating their unique, repetitive workflows step-by-step.
Steps to follow
Locked-
1
-
2
-
3
-
4
-
5
Key Points
LockedWorth watching if: Viewers interested in practical AI-driven software development will find this video highly valuable. It is most useful for those looking to move beyond simple AI coding prompts and into building robust, autonomous agentic systems. The segment on 'agent loops' and the Cloudroom ADE demonstration are particularly informative for anyone seeking to automate their own daily workflows.
Sign in to unlock the full extract
Every claim, key point, and timestamp for this David Ondrej video — plus a daily email of every channel you follow.
Sign in with GoogleNo credit card. Free tier forever.