You choose a model for the job, write a prompt that names the task, and use what comes back.
Say what prompt engineering is, explain how a model builds an answer, and tell a weak prompt from a strong one.
Say what prompt engineering is, explain how a model builds an answer, and tell a weak prompt from a strong one.
Pick between GPT, Claude, Gemini, Llama and Mistral for a given task, and say why.
Reach for zero-shot, few-shot, chain-of-thought or role prompting on purpose rather than by habit.
Compress a document into a memory file or a skill so that every line says what its source said.
Ask Claude Code for a fixed-column table from five differently written field reports, read its plan before anything is written, and keep the original files untouched.
Run a chain of agents with the model and the effort chosen for each step, and check the answer independently.
Turn a brief into a deck spine (audience, purpose, six beats, claim headlines) that a presentation tool can build from.
Write a visual brief precise enough that the first image is usable, and send it to the right specialist tool.
Hold one face across a set of images, then give it a voice.
Take a still to motion on a locked brief, and send each job to the tool that does it best.
Run voice, transcript and edit as one pipeline, and catch where a transcript lies.
Direct an agent to build a bot, keep its secret safe, publish it, and own the run path.
Turn a spoken message into a filed document by making the model fill a schema, not write the file.