On AI

A statement on AI development

We(I) use AI (specifically LLMs) in our product development. Most companies at this point use AI in their product development on some spectrum between IDE line completion and ad-hoc vibe-coding. We're on that spectrum too.

In the past 9 months (As of early Q4 2026), agentic coding has exploded and has impacted every part of the software development industry. For large established developers/companies, this is seen as a windfall for productivity. For newly established teams and products, it's met with skepticism. This is especially true in crafting software for creative endeavors. Because creativity is such a cherished human trait, anything which dilutes that, provides as an unearned shortcut, or masquerades as it, feels like it takes away from the soul of it. In our world today, creative ambition is combining with technology to allow people with no technical background to ship products that would have been far beyond their capabilities a year ago. This has led to the bottom end of the vibe-coding on that spectrum.

We(I) look at agentic development as a tool to be used in a responsible and thoughtful manner: should I vs. can I. It's used across the board for guided research, planning, development, tracking, and much more. Work that could have been done in years in a mature dev house can now be done in months with the appropriate level of rigor, or in a week of "YOLO". We(I) don't believe in YOLO. In the immortal words of the Head & Shoulders ad campaign: "You never get a second chance to make a first impression". Shoving half-baked slop out the door doesn't benefit anyone except my over-developed dopamine receptors.

How we build our "stuff"

After spending years in the industry working on teams building products and services serving billions of people we've come away with some understanding of the right way and the wrong way to design and build software.

Specs are at the heart of the development process. Coding doesn't start until a wave or release is defined with entrance and exit gates, success criteria. An overall roadmap to release drives the feature selection and ordering. Plans are built from the specs and a strict test driven development process builds on the plans to deliver the features. We have defined rigorous tiers of testing from Inner Loop testing, through regular integration, perf, and soak testing, to release testing. All builds have manual test criteria before going to beta testers to catch any last items.

Debugging sessions work like pair coding while determining where our assumptions were wrong or where the research ended up being incorrect - fix the cause, not the symptom. Quality gates and UX sit with human product management. Process, compliance, and rigor sit with human program management.

It would be faster, cheaper, and easier to not do any of this. It would also be the best way to set us up for failure.

Why we build our "stuff"

For the most part our(my) projects are built out of a need where I haven't been able to find a suitable alternative. The need is followed with market research: is there any value in this for anyone else? Of all those needs, how do they get stack ranked? What will other people get excited about? What should be put back out as open source? What has unique value that I can justify asking money to cover the work that was done? At the end of the day, is it something that I wish I had myself? Is it something I'd buy from someone else if it existed?

Why this statement? (an excerpt from an email thread with another developer)

I feel the transparency item seems like it'll be a growing need in building software for creatives. Because creativity is such a cherished human component, anything which dilutes that, provides as an unearned shortcut, or masquerades as it takes away from the soul of it. I don't know where to draw the line in the spectrum between corporations like AVID or Adobe incorporating AI/LLM-assisted coding and a person in their bedroom "one-shotting" their fever-dream monster synth. Maybe it's in the creation of something novel, or a fundamentally new approach vs. "Make me Serum but better". Does an established developer get a pass once they've paid their dues? Is there greater value in software (like any crafting) which is coded by hand? Circling back to transparency, I guess we're just in a world where we need to [detail our inner workings] (or make a 35m in-depth YouTube video about the research process) until.... until what?

Until we get to a state where development with AI becomes normalized and accepted across the creative software industry, it's in mine and the user's best interest to be transparent on how we(I) work.