A tool list tells you what exists. A workflow tells you how the pieces fit — which tool at which step, what it costs to run once, and where it falls apart.
“When you ask the same questions of everything, you begin to reveal the differences between them.”
Specify it, make the assistant plan it, build the smallest version that runs, then find and describe exactly what is broken. The defect log is the skill; the artifact is what you have left at the end.
Open the steps for Building a working thing you could not build alone, then making it live →You already have ten years of enrollment, scheduling and outcome records. The problem was never collecting them. It was asking them anything.
Open the steps for Making a decade of program data answer one question →Your dashboard knows the fill rate. It does not know that the room was freezing and half the class had no textbook. Put both in the same table.
Open the steps for Keeping field notes next to the numbers →Years of outside feedback contain a signature you cannot see from inside. Pull the recurring vocabulary, cut it to six words, and use it.
Open the steps for Finding the handful of words that define what you do →Getting quantities off a drawing set fast is the point. Being able to show where every number came from is what keeps the bid from becoming a loss.
Open the steps for A quantity takeoff you can defend line by line →You have explained the same temporary layout three times and the electrician still built it wrong. Draw it, hand it over, and watch them read it.
Open the steps for Drawing the install so the trade can actually build it →A phone photo in the field gets you a differential in a minute. The skill being taught is the escalation rule, not the identification.
Open the steps for Diagnosing a crop problem from a photo, and knowing when to escalate →Before a program buys the proprietary box, spend an hour learning how it works. Sometimes you still buy it, and now you know what you are buying.
Open the steps for Understanding the expensive system before you buy it →New drivetrain, new surface, new system. Build the mental model first, check it against the service information, then go slow.
Open the steps for Briefing yourself on an unfamiliar platform before you touch it →Arriving at counsel with a structured draft and a list of open questions costs less and teaches more than arriving with nothing — or with a signed template.
Open the steps for A first-draft agreement that saves the lawyer time instead of replacing them →Generate sixty directions by Tuesday, then rebuild the chosen mark by hand on a construction grid — because a PNG is not a logo.
Open the steps for Taking a brand identity from written brief to editable vector deliverables →Separate recovering information that survives in the negative from fabricating information that does not — then label the difference where the viewer can see it.
Open the steps for Restoring a damaged archival photograph, and disclosing what the machine invented →Captions, transcript and audio description done to professional standard — where writing description for your own film turns out to be an editing exercise that changes how students shoot.
Open the steps for Making a student film fully accessible on a real deadline →Build a LoRA from your own drawings, discover what it captures and what it flattens, then make and defend an authorship claim about what comes out.
Open the steps for Training a style model on your own archive, and arguing for your authorship of the output →Peer review first, AI second, and the graded artifact is the reflection comparing them — which is where the thinking actually shows.
Open the steps for Getting every student substantive feedback on a draft when you have ninety of them →Structured role-play with a voice assistant between class meetings, where the graded artifact is a marked communication breakdown, not the transcript.
Open the steps for Twenty minutes of speaking practice a week that a twenty-eight-person class cannot provide →Hand students a polished, quietly wrong AI lesson plan and make the assignment finding what is wrong and naming which principle it violates.
Open the steps for Grading the machine's lesson plan: building developmental judgment by critiquing AI activities →AI drafts descriptions and captions at volume while the judgment — what the image is for — stays with you, because that is the part the machine genuinely cannot do.
Open the steps for Remediating a semester of course materials without hand-writing two hundred alt texts →Timestamp a ranked hypothesis list before any lookup, then write up where the repair-information system disagreed with you and why.
Open the steps for Commit to a diagnosis before you open the database — then defend the delta →The graded artifact is the documentation, not the calibration — and the AI's role is an adversarial completeness check plus a live demonstration of why you never trust it with a specification.
Open the steps for Build the ADAS calibration documentation package a plaintiff's attorney would subpoena →Students get the problem as a picture with no text, write the spec, the model generates from it — and the spec is what gets graded. Then they hunt a bug in code that passes some tests.
Open the steps for Grade the specification, not the program: prompt-first CS1 with a refutation round →Require the model to do the computation so the entire grade can sit on the assumption audit, the design critique, and the interpretation in context.
Open the steps for The AI runs the test, the student defends the assumptions →Every claim in the final memo has to point to a line in a real transcript with a real person's initials and a date.
Open the steps for Customer discovery: AI writes the interview guide, real humans give the findings →A mock intake, an AI summary, and then the part nobody teaches: reconciling it against what was actually said and flagging issues without answering them.
Open the steps for The intake summary that stays on the paralegal side of the line →Baseline where a client appears across assistants, diagnose why competitors get cited instead, ship one change, and re-measure the same way.
Open the steps for Does the AI recommend our client? An AI-visibility audit for a real local business →Build the playbook first, review a real contract against it, then defend the redlines — because the skill is materiality judgment, not clause spotting.
Open the steps for Redlining against a playbook: when the AI's issue is not your client's issue →Draft the beats in a chatbot, lock a visual style, generate frames that hold the same characters across shots, then cut a timed animatic — without anyone needing to draw.
Open the steps for Turning a written shot list into a style-consistent storyboard →Generate genuine reading-level variants of a single text so a mixed-proficiency section can discuss the same content — then check what the simplification quietly threw away.
Open the steps for One reading, four proficiency levels, same discussion →Run a research question through an AI tool, then pull and confirm every single authority by hand — the step whose omission produced over two thousand sanctioned filings.
Open the steps for The verification lab: catching AI-fabricated authority before it reaches a filing →Run your own homework through the tools students use, find what survives, and rebuild the assessment around the problems that do.
Open the steps for Auditing your problem set against the solvers →Use AI to assemble a competitor and market picture fast, then make the verification of each figure the graded work.
Open the steps for A market analysis where every number gets traced back →