Using Code Agents Effectively

Every time LLMs get an "upgrade" they feel like magic, the first impressions are always so strong. That's how I felt with code agents, all of a sudden models were hyper context aware to my exact system, and could iterate on problems until they made a working solution. The craziest part is I don't even have to understand the solution because the Agent already implemented it. But as it always goes, playing with this new capability you quickly find it has hard limits. Beyond creating a toy project they begin to cut major corners, or making any edits to a large repo they just completely fail. The latest models, GPT5-Codex and Sonnet 4.5, are well into the territory of being as capable as a new engineer. However, they need to be directed with extreme nuance and can't just be let loose on problems like an eager new grad. The models are smart enough to get just about anything done, but it not without a lot of teamwork. Below I'll detail some of the strategies I'm using to get good results wit