Burke Holland, a GitHub Copilot engineer who runs AI agents daily, posted in early January that Claude Opus 4.5 represented a genuine step-function capability jump, not an incremental update. This conversation with Adam Stacoviak unpacks exactly what that means in practice, grounded in hands-on experience from someone building with these models professionally.
The episode covers Opus 4.5 in detail but does not stop there. GPT-5.3 Codex enters the conversation as a direct comparison point, and the show notes reference Cloudflare rebuilding Next.js with AI in a single week, a data point that reframes what "capable" even means right now. The 17-minute Changelog++ bonus segment suggests the most candid material is behind the paywall.
The reason to read Holland's original post and then listen to this episode is the methodology, not just the verdict. Holland works on Copilot, which means his benchmark is not a toy project. The show notes also surface Entire, a new developer platform from former GitHub CEO Thomas Dohmke, which adds competitive context worth tracking.
[READ ORIGINAL →]