Web Directions Year round learning for product, design and engineering professionals

Fully Automated Luxury Gay Space Engineering: What AI-First Software Development Actually Looks Like

The title is playful, but the question underneath is serious: what happens when you actually commit to AI-first software development, not as a buzzword but as a genuine architectural principle?

Most organisations that talk about "AI-assisted development" still operate the same way they always have. Engineers write the majority of the code; AI fills in the gaps. But what if you inverted that? What if AI handled the bulk of code generation and engineers focused on the parts of software creation that require judgment: architecture decisions, code review, quality assurance, and design trade-offs?

That's the experiment Daniel Rodgers-Pryor has been running at Stile AI Labs, building educational products at speed. The results are instructive because they reveal both what works surprisingly well and where the approach hits real friction.

The case for this inversion is compelling on the surface. Code generation is mechanical. Architects are expensive. If AI can handle the mechanical parts reliably, engineers can focus on the parts that require human insight. The promise is faster iteration, less routine work, and product teams that ship more frequently.

But the reality is more nuanced. AI-first development creates new problems: code quality risks that are harder to spot because there's more of it, integration challenges when generated code meets legacy systems, and a different kind of cognitive load on reviewers who now need to verify correctness across much larger changes. There are also subtle cultural shifts. When engineers aren't writing most of the code, the relationship between the person and the product changes.

What makes Rodgers-Pryor's work valuable is that it's candid about the tradeoffs. Which patterns of AI generation actually produce reliable code? Where do you still need human expertise from the start? How do you structure teams and review processes to catch the failures that AI-first development is most likely to miss? These are the questions that separate genuine engineering insights from promotional thinking.

This matters now because the economics are shifting. If teams can generate significantly more code with fewer engineers, that changes staffing, hiring, and skill requirements across the industry. But only if you can actually make it work reliably. Learning what works and what doesn't — not in theory but in real products used by real students — is the kind of practical knowledge that's valuable now.

The answers aren't obvious. They require experimentation, honest assessment of failures, and willingness to reject approaches that look good on paper but don't work in practice. That's what makes this conversation worth having.

Daniel Rodgers-Pryor, Head of Stile AI Labs at Stile Education, is presenting this talk at AI Engineer Melbourne 2026 on June 3-4.