Claude Sonnet 5 shipped today, and I'm already rewriting tomorrow's lesson plan. I teach non-technical people how to use AI for their actual jobs — contracts, client emails, financial models, marketing plans. Not coding demos. Real work. Here's what's different with this release: it holds context longer and reasons through multi-step tasks without losing the thread. That sounds small. It isn't. Most of my students don't need a chatbot that's clever for one turn. They need something that can take a messy brief, ask the right follow-up, and carry that understanding through ten more steps without forgetting what they told it five minutes ago. That's the actual bottleneck in knowledge work right now — not "can AI write," but "can AI stay coherent long enough to be trusted with something that matters." I haven't run it through a real client project yet, so I won't pretend I have a verdict. But the early sessions feel less like prompting a tool and more like briefing a sharp colleague who doesn't need everything repeated. If that holds up, the skill that matters next isn't prompt engineering. It's knowing what to delegate and what to check. What are you noticing in your first day with it?