Each one below started as a concept from an Anthropic course and became something you can actually touch — try the mini version right here, or open the full tool. New ones get added as I build them.
Built with Claude — every tool here was designed and coded in collaboration with ClaudeAnthropic Academy — anthropic.skilljar.com
Handing a task to AI works a lot like delegating to a new hire: you'd decide upfront what they handle alone, what needs your sign-off, and what you're accountable for either way. This framework turns that instinct into four short questions you answer before you ever open a chat.
Answer one question below, and see how it becomes the start of a real, usable prompt.
The live sibling of 4D Prompt Builder — brings your own Gemini key to a twelve-question Socratic interview that corrects vague or repeated answers as you go, mirrors Description against Discernment, scans five cross-dimension gaps, and scores how solid each D actually is.
Open the full architect →A different job for the same four D's — this one isn't for getting work done, it's for making sure AI helps you actually learn. Walk through learning-mode Delegation, Description, Discernment and Diligence, pick your study protocols, and get a reusable instruction block to paste into any AI chat before you study.
Open the full configurator →The full-scale version of the same idea: instead of one prompt, four stages of AI collaboration with a real human decision at the end of each. Discover roles worth considering, prove your fit in a graded case study and mock interview, then get an honest final recommendation grounded in evidence from your own answers — never a default to what you said you wanted at the start.
Open the full companion →Transparency about AI use is becoming a career skill of its own. This tool interviews you — never a form — about what you actually asked AI to do and what you did yourself, flags language that inflates either side, then produces honest disclosure statements plus a formal verification letter an examiner or employer can act on.
Open Disclosure Studio →Next tool in this topic
Coming soonAnthropic Academy — anthropic.skilljar.com
From Markov chains guessing off the last word alone, to transformers weighing everything you've written so far — the method changed, but the game stayed the same: predict what comes next, one token at a time. Seeing how each generation actually guesses makes it obvious why fabrication concentrates exactly where specifics run out.
Pick an algorithm generation, and see how differently it guesses the next word.
Push one dial hard, and see what it does to the resulting prompt.
Pick a metric, and see how it judges "The car is fast" vs. "The vehicle is quick" — zero shared words, same meaning.
Drag a session-length slider and watch a real blog paragraph quietly reword itself — no warning shown, ever.
A real ~2,500-word test document, a buried fact, and two lookalike decoys — see if an AI grabs the right one.
Got a document too long to trust an AI with in one go? Split it into small overlapping pieces, extract each one safely, then stitch the results back together — with a check for what survived.
Paste a draft prompt and see, line by line, which instructions are checkable and which are just a hope — then fix the vague ones and build a refined prompt to actually use.
One topic, four dials — next-token style, knowledge depth, working memory, steerability. Move a dial, answer a question, and see how the actual finished writing changes.
Paste an AI answer and run it through five known compound-failure patterns — every time, whether anything feels wrong or not, because fluency and correctness aren't the same signal.
Not a blog tool, and it can't make AI more accurate — it maps your task onto all four properties and tells you exactly where to spend your verification effort.
Next tool
Coming soonGeneral AI-fluency practice — applying the same habits outside any one course module.
Anyone can write a skill on a resume. What actually convinces someone is a specific, checkable example — something they can click on. This tool walks a claim through four steps: sharpen it into one honest sentence, plan where the proof will live, then verify it's the kind of thing real people would actually confirm.
Pick a skill you'd claim, and see it turned into the start of a proof-hunting prompt.
Pick an opening line, and see one of the mechanical checks it runs before you post.
Pick a question, and see the kind of gap it would flag before rewriting it.
Pick a paragraph pair, and see the kind of relationship it would flag before any fact-check happens.
Pick a paragraph type, and see whether it even belongs in the fact-check queue.
Next tool in this topic
Coming soon