AI & projects
Coding agents · Public work
How I use coding agents
Coding agents are part of my daily work. I use them for research, implementation, tests, review, debugging and documentation. I often have several agents working in parallel while I concentrate on the decisions that need my attention.
The context I give an agent matters. A useful assignment includes the intended outcome, the relevant systems, constraints, supporting material and a way to establish whether the work is complete. I review the proposed approach and the result, question assumptions and redirect the work when it misses the point.
I remain responsible for what is built and what ships. I want evidence that the result works in its intended setting, including the user flow when the change affects one. A successful command or a passing test can answer a specific question; I still need to understand what it establishes about the actual product.
A public project: Creative AI Workflow Lab
Creative AI Workflow Lab ↗ is a local prototype of an AI-assisted design workflow. A Figma plugin connects to a backend for copy, localization and image workflows, with a browser interface for exercising the same flow outside Figma.
The project explores model orchestration, policy, explicit review and application of results, usage records and verification. Model and image outputs use deterministic fixtures rather than live generation. The repository includes the prototype, architecture notes, trade-offs, a delivery plan and executable verification scripts.