Burke Holland, a GitHub Copilot engineer who runs AI agents daily, went public in early January with a direct claim: Opus 4.5 is a step function in capability, not an incremental update. This episode is his technical debrief with Adam Stacoviak, and it covers the specific reasons why, grounded in Holland's hands-on work building with the model.
The conversation does not stop at Opus 4.5. GPT-5.3 Codex enters the discussion as a legitimate competitor living up to its own hype, and the episode connects to a broader cluster of tools and projects worth tracking: Cloudflare's one-week AI-assisted Next.js rebuild, Thomas Dohmke's post-GitHub CEO venture Entire, and Layercode's voice AI infrastructure for TypeScript developers. The show notes alone are a dense reading list.
The reason to read the full transcript rather than stop here: Holland's framing of what specifically changed in Opus 4.5, beyond benchmark scores, is the core argument. Changelog++ members also get 17 additional minutes not available in the public feed.
[READ ORIGINAL →]