Six skills, one path: pick the right AI tool, prompt it well, fix output that looks like garbage, use AI without going soft, learn fast with NotebookLM, and stand up a real Claude Cowork assistant that runs a job on its own.
Most people have opened ChatGPT or Claude, typed something in, gotten a mediocre answer back, and quietly decided AI is fine but not that big a deal. The real gap isn't the model. It's six specific skills nobody teaches: picking the right tool for the job, writing a prompt that actually works, knowing why the output disappointed you, using AI without going soft yourself, using a tool built for grounded research instead of guessing, and setting up an assistant that does real work instead of just answering questions.
This path builds all six, in order. You'll start with a three-question method for matching any task to the right assistant — ChatGPT, Claude, Gemini, or Perplexity — instead of guessing or defaulting to whichever one you already have open. Then you'll build prompts with a simple five-box structure that turns vague requests into results you can actually use, and assemble your own working prompt as you go.
From there you'll diagnose the three real reasons AI output looks like garbage — a free model doing the work, context you never gave it, a task that was vague from the start — and fix each one. You'll also learn the trade-off nobody talks about: how to get AI's leverage without quietly losing the skill it's replacing, and how to draw that line on purpose instead of by accident.
Then you'll pick up NotebookLM as a grounded research tool that only speaks from your own sources, so you can learn something fast without the hallucination risk of a general chatbot. And you'll finish by standing up a real Claude Cowork assistant — scoped to one folder, bound by rules, loaded with your context — running one supervised, recurring job that keeps working every week without you rebuilding it from scratch.
By the end you're not someone who "uses AI" in a vague sense. You have a working method for choosing tools, a prompt structure you reuse on every task, a diagnostic for bad output, a rule for staying sharp instead of soft, a research tool that won't lie to you, and a standing assistant already doing one real job on its own.
The casual user who wants results, not more experiments: already has ChatGPT or Claude open most days, but every session starts from zero and every output needs heavy editing. Wants a repeatable system instead of trial and error.
The AI-skeptical operator: has been burned by generic, hedge-everything, or confidently-wrong output enough times to wonder if the tool is overhyped. Usually it's the setup, not the model — this fixes the setup.
The owner or team lead ready for a real assistant: wants AI handling an actual recurring job — a weekly brief, a research folder, a status update — instead of one-off prompts that need babysitting every single time.
A three-question method for matching any task to the right AI tool, plus an honest, dated read on what each one is currently best at.
A five-box prompt structure you can reuse on any task, built from a working prompt you assembled yourself.
A diagnostic for the three real reasons AI output disappoints — and the fix for each one.
A clear rule for using AI without losing the underlying skill it's replacing.
A NotebookLM workflow for learning anything fast from your own sources, without guessing what's true.
A working Claude Cowork assistant — scoped, rule-bound, loaded with your context — already running one real supervised job that repeats on its own.
This course is part of the academy's invite-only offering.
Access is granted by the academy after you join.
