July 2025

Review and hands-on of Charm Crush, a sleek CLI AI coding assistant with standout UI, solid agent performance, helpful summaries and model switching, but hampered by repetitive permission prompts and occasional hangs.

Talks and live demos showcase building AI agents, agent SDKs, autonomous browser companions, and secure private‑data workflows on Convex and beyond.

Alex stress-tests the GEEKOM IT15 mini PC—benchmarks, ML training and LLM inference, battery-powered runs, upgrades to 96GB RAM, and brutal physical abuse—showing it still performs and can run sizable models via Vulkan on the iGPU.

A hands-on evaluation finds Qwen 3 Coder performs close to Claude 4 Sonnet across multiple IDE agents, while noting key drawbacks around missing vision, configuration sensitivity, and potentially higher costs depending on provider and caching.

Cody maps a realistic hype spectrum for AI in programming, sharing when it boosts small-project productivity and where it falls short on complex, enterprise-scale codebases.