Hangar DX Podcast

The episode contains several solid conceptual insights - particularly the reframing of code review as fundamentally about taste and organizational judgment rather than bug-catching, and the notion that falling costs enable collapsing PRDs/architecture docs into the PR itself. However, these insights are interspersed with substantial throat-clearing, exploratory thinking, and restated premises that dilute density. The discussion of potential 20x unmerged PRs is provocative but underdeveloped; the spec-driven development tangent meanders without concrete takeaways.

“code review process is fundamentally built around applying judgment and saying hey is this the right thing for my product?”

“as it's becoming cheaper to write and rewrite code, a lot of the things that we used to do that we used to front load, uh, in order to make the same types of decisions before code existed, uh, are kind of collapsing into the same space”

The guest offers a genuine reframing - code review as taste/judgment rather than bug-catching is a useful counterpoint to mainstream discourse. The idea of collapsing pre-code decision-making (PRDs, specs, docs) into the PR context due to reduced costs is fresher than typical takes. However, the underlying concepts (collaboration, verification, knowledge sharing as review functions) are well-established. The discussion of AI and LLMs in this context is exploratory rather than novel; the guest explicitly acknowledges uncertainty ('we're all kind of inventing it as we go').

“code reviews were about more than just is the code correct...so much of the code review process is fundamentally built around applying taste”

“there's a whole slew of things that happen at this point...much more uh, taste and applying organizational processes”

David Poll is a Senior Director of Engineering at GitHub with a decade at Google (Firebase, Cloud, Android DX) and prior startup experience at Parse. He has direct responsibility for GitHub's code review and pull request product surface, making him exceptionally well-positioned to speak on this topic. His background spans both large-scale engineering culture (Google) and product building for developers. This is a genuinely senior, relevant operator who has shaped the tooling and practices around the very subject being discussed.

“I am a senior director of engineering at GitHub...responsible for the code and review organization which has uh, as part of it, pull requests and the pull request experience and then Copilot, uh, Code Review”

“I've spent my career though building stuff for developers...came by way of uh, Google for a decade”

The episode is notably thin on named examples, concrete metrics, and specific case studies. The guest mentions Firebase, Parse, Google, Cloud SDK, Excel, Netflix, and Silverlight in passing, but rarely ties them to concrete data or learnings. The '20x unmerged PRs' thought experiment is never grounded in actual numbers or projects. Most claims remain at the conceptual level ('there's a lot of that happening'), lacking quantified evidence or named companies/products demonstrating the patterns discussed.

“I was actually a PM at Microsoft at the very beginning of my career working on Silverlight”

“there's a humongous portion of software development happens...these are onesie twosie things”

The host asks reasonable opening questions and does attempt to probe (asking about specs in AI-driven development, pushing back on the pre-code-review era claim). However, follow-ups are often soft or exploratory rather than sharp. When the guest makes vague claims ('I think there's a whole bunch of it that is...'), the host rarely drills down for specifics. The conversation meanders into tangents (Silverlight, Excel, Netflix infrastructure) without sharp redirection. The host also agrees with the guest frequently ('I feel like this is also sort of like we are getting there'), limiting productive tension.

“Can there uh, be enough trust built into the AI systems?”

“I'm going to challenge something you said a little bit”