I use AI-assisted tools where they are useful, including code generation, implementation, refactoring, research, planning, and development support. I do not treat AI output as an engineering decision, a design decision, or a substitute for domain knowledge.
I care less about who typed each line of code than about whether I understand it, verify it, and take responsibility for shipping it.
I decide and evaluate product structure, signal flow, DSP behavior, parameter design, performance targets, and release criteria. I make audio decisions using domain knowledge built through my work with Max/MSP, TouchDesigner, and analog and digital audio systems.
I may use AI to generate or revise code, boilerplate, tests, documentation, and implementation alternatives. I review the output, modify it where necessary, and validate it against the intended architecture and behavior rather than accepting it simply because it compiles.
I combine automated checks with manual debugging, listening tests, real-host testing, state-recall checks, CPU profiling, and repeated QA on actual operating systems and DAWs. I do not treat automated test success alone as release approval.
I give server-connected components additional review, including static analysis, dynamic testing, input and error-path testing, and manual QA. I do not trust AI-generated security-sensitive code without independent validation.
I generally begin interfaces with my own sketches, information hierarchy, and interaction decisions. I may use AI tools to explore alternatives, organize ideas, or assist with planning. I refine the final hierarchy, layout, assets, interaction details, and visual polish manually in tools such as Figma, Photoshop, and Illustrator. I do not treat AI-generated visual output as finished product design.
I define the product, signal flow, user interaction, performance requirements, and what should not be compromised.
I use AI where it speeds up research, iteration, planning, implementation, boilerplate, or refactoring.
I check behavior against DSP, realtime, UX, compatibility, and security requirements rather than accepting generated output at face value.
I test builds in real hosts and environments, reproduce failures, profile bottlenecks, and fix issues before release.
I make the final decision to ship, and I take responsibility for what ships—not the tools I use during development.
The tools I use may change as development tools evolve. My principle remains the same: AI assistance does not replace my engineering judgment, domain knowledge, security review, design direction, testing, or accountability.
Last updated: September 5, 2026
