If you use Claude mostly as a chat or coding helper, this is the part worth looking at before you celebrate another big AI launch. The easy mistake is to treat Claude Code and Claude as the same tool, then read every announcement as a pure capability upgrade. That is how you end up optimizing the wrong skill and wondering later why the shiny new release still does not help you ship.
The post this is based on is narrow and specific: Anthropic officially launched 13+ FREE AI courses with certificates, including Agentic AI and Claude Code [C001]. The interesting part is not just the count. It is what Anthropic chose to put in the curriculum spine.
On Anthropic's public learning pages, prompt engineering is not presented as the whole game anymore. The course catalog places Claude Code, the Claude API, MCP, tool use, RAG, and broader learning tracks beside it. Inside the 'Building with the Claude API' course, prompt engineering and evaluation is only one part of a bigger build path that also covers tool use, retrieval, MCP, Claude Code, computer use, agents, and workflows. On the 'Build with Claude' page, agents, skills, MCP, tool use, and evaluations are grouped as parallel topics instead of extras.
That is the real tell. The part people argue about is never just that the model got better. It is why the strongest thing was not put on the table as the whole story. My read is simple: if you can't work with tools, being great at prompts still won't get useful work shipped [C002]. Prompting is not useless. It is just no longer enough when the work touches files, data, apps, or anything you want to repeat without babysitting.
There is a boundary here. This is not proof that prompts are dead. It is not a live model test, a benchmark comparison, or a product teardown. It only supports a narrower judgment based on Anthropic's public course and learning pages: prompt-only learning is being demoted relative to API, tool use, RAG, MCP, Claude Code, and workflow skills. At least from that framing, Claude Code looks better suited to help you inspect and structure the problem first, while Claude looks better suited to carry more of the later work through.
If you know someone still collecting prompt templates but has not connected a model to a single tool, share this with them.