There's no new language for AI yet, and probably won't be one that looks like Python or C. The skill is context engineering: what you include, what you leave out, and the order it comes in.
Context engineering, not grammar
I asked Qwen what it takes to communicate well with AI. Plain English is still the interface. What's forming around it is structure.
XML-like tags such as context, task and constraints turn up everywhere. The Model Context Protocol is a standard for connecting AI to files, APIs and databases. Chaining breaks a task into steps and feeds outputs between them. Spec-style prompting writes detailed requirements the model can work against.
The useful part was simple. Say what you're trying to do, give the context, and be clear about the output you need. Turning a fuzzy idea into something useful is a skill of its own.
- Write briefs, not questions: give the audience, the goal, the tone, the constraints and what to avoid.
- Build a prompt from role, one clear task, constraints, an example and the exact output format.
- Refine in two or three rounds instead of expecting the first prompt to be perfect.
- Give feedback like a manager: name the tone problem and the direction, like shortening it and adding a personal opener.
Tools and links
- QwenThe model I asked what it takes to communicate well with AI.
