Engineering workflow
How I build with AI
I build with AI agents by giving them clear outcomes, relevant context and defined boundaries. I combine community-built capabilities with custom skills, then use testing and human review to evaluate the result.
The workflow
Define the outcome
Clarify the problem, constraints and acceptance criteria before implementation.
Prepare the context
Provide the relevant project conventions, documentation and task-specific instructions.
Choose and adapt skills
Use community-built capabilities where they fit, and custom skills where the work needs project-specific knowledge.
Implement in reviewable steps
Use agents to investigate, prototype and implement changes in manageable increments.
Verify and iterate
Run the relevant checks, inspect failures and refine the implementation. I’m developing additional verification skills to make this feedback loop more repeatable.
Review before release
Review the code and user experience, check unresolved risks and decide whether the change is ready to ship.
What I build myself
I combine community-built skills and harnesses with custom instructions and workflows. The custom work captures project context, task boundaries and repeatable engineering steps. My focus is making agent-assisted development useful within a real software project.
What stays private
Some of my custom skills and project-specific workflows are private. This page explains my approach without publishing internal code, proprietary instructions or sensitive project data.
Where human judgement matters
Agents help with exploration and execution. I remain responsible for understanding the problem, assessing trade-offs, reviewing changes and deciding what is ready to ship.