Kimi K3 Reaches the Coding Frontier. Here's the Catch
Kimi K3 is genuinely near the best closed coding agents, but it gets there with heavy trajectories, cache-dependent pricing, and a different task profile from GPT-5.6 Sol.
George Lydakis on AI agents, evaluation, and developer tooling — plus the open-source projects behind the writing.
Kimi K3 is genuinely near the best closed coding agents, but it gets there with heavy trajectories, cache-dependent pricing, and a different task profile from GPT-5.6 Sol.
The universe always moves toward higher entropy, and AI looks like the next big engine that accelerates that flow through civilization.
There are three problems with AI agents right now that are deeply coupled — the human bottleneck, the quality ceiling, and the cost. They reinforce each other in ways that make the "AI will 10x everyone" narrative more fragile than it sounds.
Jul closes the agent coding loop with continuous sync, automated CI/review feedback, and provenance traces.
It used to be that tech was something niche. That people interested in gadgets like that were geeks and powerusers; and as such anything new was treated with excitement and wonder first and any skepticism was secondary.
Last Monday Apple announced their entry into mixed reality, or what they call Spatial Computing. Having watched the presentation and followed the subsequent coverage, I have a few insights and observations I'd like to share.
Welcome to Engineered Out of Office! I'm George, and this is a new space for me to put out my thoughts, things that I'm excited about, and talk about projects I'm working on. This blog itself came out of the desire to build something outside the confines of work; to create something for the sake of creating and having fun.