Judge what the AI did. Not just what it said.
Your manager wants the AI assistant to do more. Decide how much autonomy the job actually needs before you hand any of it over.
Two runs of an AI assistant can write the same excellent customer update while one of them opened records it had no business in and sent without approval. Learn to size a workflow's autonomy, fence its permissions, test it, read its action trace, set its stop conditions and decide what it needs after a change.
Training + AI coaching + reusable workflows. Learn to supervise AI work and use Coach Q to prepare operating instructions, permissions and test cases. Your team tests the workflow and decides what it may do.
Built by a quality auditor with 20 years in the industry, who reads an agent's action trace the way an auditor reads a record.
100+ years of combined quality and compliance experience, boiled down to one thing: knowing when to trust the AI, and when to stop it.
See how it works ↓Try lesson 1 before you pay. 3-day refund after you buy.
One-time course purchase. Lifetime lessons and your AI coach included.
Coach Q, your private AI coach, comes with your course. No subscription or renewal fee. Your purchased lessons remain available for life.
Training a team? Everyone who works with AI learns where to check it, so mistakes get caught before they reach a customer.
10% off for 5 or more people.
- One operating brief for an AI workflow at your job: permissions set, tests written, traces read, stop conditions in place
- A certificate anyone can verify, after an unseen case reviewed by a person and a 30-question final exam
- Private AI coach for this subject, included with your course
- Six hands-on labs that build the brief, one piece at a time
- Every lesson, yours for life
3-day refund window. Refund policy
More than 50 people, or need an invoice instead? Email will@qmslearning.com for a quote.
Learn it. Practice it. Bring it to work.
Build subject knowledge, earn your course certificate and prepare work you can check and use.
Pass an unseen case reviewed by a person and the final exam, and your certificate records exactly that. Anyone can confirm it at academy.qmslearning.com/verify, and you can add it to your LinkedIn profile in one click. It records your course achievement; your workplace results come from applying and checking the work.
Work through the lessons at your own pace. Ask Coach Q to explain a concept, try another example or question your reasoning. Use the course study plan and focus timer to fit learning around your week.
Use Course Hub to create workflow materials, save revisions and keep reusable templates. Download your work for review by the people who will use it. Your lessons remain available for life, and AI coaching comes with your course.
Learn the skill. Put your AI coach to work.
When your team wants to let an agent send customer emails on its own, return to the method with Coach Q. Your AI quality and compliance colleague helps you think through the problem and build a workflow you can review.
Know what good work needs
Learn supervising AI agents, work through the course scenarios and ask your AI coach to explain what is unclear. Bring your own reasoning and ask it to challenge the gaps.
Build a reusable workflow
Use the Course Hub document tools to prepare an agent operating brief, a permission matrix, a six-test set and an exception playbook, using the facts you supply.
This whole course is the check: you'll read an agent's action trace, find where it went past its permissions, and decide what goes back to a person.
Start the next job with a method
Edit your workflow materials, save revisions, download your work, or save the structure as a reusable template. For the next case, replace the old facts and check the new evidence before drawing conclusions.
Your saved work is in Course Hub when you return to the same account.
An expert in your pocket means an AI coach within reach. Learn in your browser, then return when you need help applying the subject. AI output can be wrong; you and your organization own the review and the decision.
Coach Q, your private AI coach, comes with your course. No subscription or renewal fee. Your purchased lessons remain available for life.
Be the person who knows when to say yes, and when to say stop.
Short lessons, a plan that fits your week, and a coach who knows the job. These are real screens from the course, and the label says when one comes from another course. Click through at your own pace.
Practice catching AI mistakes before they reach a customer.
Short lessons, then hands-on exercises built on situations you'll actually face. Make your call, then see how an experienced pro would handle it.
- Built on situations you'll actually face
- See how a pro would handle it
- Pick up right where you left off
Ready when you are.
$129 per person, one-time. Lifetime access to the lessons · AI coach included · No subscription · 3-day refund window.
10% off for 5 or more people.
Someone wants to switch on automatic sending. You're the one who has to say yes.
“The demo looked perfect. Let's turn it on.”
One impressive demo tests one ordinary case. You'll build a test set that covers the missing, the conflicting, the duplicate, the hostile and the broken, before anyone trusts the workflow.
“The assistant says it updated the record.”
The final message is the workflow's account of its own work. The trace is the record. You'll score runs on what they did, and separate attempted from completed from confirmed.
“Rosa approves the session every morning, so we're covered.”
Approval at the start of a session doesn't cover every send after it. You'll put the approval on the consequential action, where the consequence is.
“It only has read access. What could go wrong?”
The right to read isn't the right to change, and neither is the right to open another client's folder. You'll write permission lines that name records, actions, recipients, limits and approval points.
“It timed out, so it tried again. Now the customer has two emails.”
Retrying a read changes nothing. Repeating a send doubles the effect. You'll set bounded retries per action and exits that go to a named owner.
“We only swapped the model. The prompt didn't change.”
A new model or a new tool permission can change behavior on its own. You'll decide when to retest and choose release, restricted trial or rollback from the evidence.
Myth or fact?
Four things people say about supervising AI workflows. Make your call on each one.
“If the final message is correct, the run was fine.”
“Approving the session each morning counts as approval for every send that follows.”
“An attempt to open an off-limits record fails the boundary test even when another control blocked it.”
“Changing the model can change the workflow's behavior even when the prompt stays the same.”
0 of 4 checked. Tap Myth or Fact on each one.
Six modules. Six labs. One operating brief.
Counted from the live course.
You don't just read about supervision. You write the brief.
Each lab adds one part to the same operating brief for a fictional assistant. No live accounts and no real messages, so you can make the hard calls safely.
An autonomy decision you can defend
Fixed rule, reviewed AI output or agent: the simplest design that does the job, and a reason for any autonomy you add.
Module 1A permission matrix with a real approval point
What the workflow may read, change and send, which limits the platform enforces, and where a named person approves.
Module 2A six-test evaluation set
Normal, missing, conflicting, duplicate, malicious and tool-failure cases, each with an expected action, a prohibited action and a pass condition.
Module 3Runs scored from their traces
Three traces read event by event, with attempted, completed and confirmed kept apart, and unsupported success claims called out.
Module 4An exception playbook
Bounded retries per action, stop conditions, and exits that tell a named owner what's known, what's unknown and what wasn't done.
Module 5A monitoring and change checklist
Owner, review sample, failure measures, version record and retest triggers, plus a release, restricted trial or rollback recommendation after a change.
Module 6Learn it, practice it, prove it, use it.
A short guided lesson per module
Six modules, each a guided lesson followed by a lab. Every activity has a click, keyboard and text route. No countdowns.
Labs on a fictional assistant
Make the calls in each lab and see why. Hints get more specific each time you ask, every hint is logged with your work, and you can retry as often as you like.
An unseen case and a final exam
Evaluate a different assistant in a new case: a capstone scored out of 20 and six decisions scored out of 12, reviewed by a person. Then a 30-question final exam, 80 percent to pass.
Your own workflow
Leave with a toolkit of the questions to ask before any AI workflow is switched on, and a plan for the first one you'll supervise. Your coach comes with your course.
Ask the question your team is about to ask. Get a straight answer.
Coach Q is on every screen. It asks what evidence you used, points you at the trace or the policy, and hints without writing your answer. You make the call.
Not for the sends. Approving the session covers starting the session. The consequential action is the send to a customer, so that's where a named person approves.
And a prompt isn't a lock. If the platform can hold a message until someone releases it, enforce it there, and keep the instruction as a second layer.
Which actions in your workflow could reach a customer or change a record? Those are the ones to list first.
From your first lesson to an AI workflow you can sign off on.
6 modules. About 3.5 hours of self-paced work.
- Module 1
Choose an appropriate amount of autonomy
Pick the simplest design that meets the task, and justify any autonomy you add. 2 lessons.
- Module 2
Define tools, permissions and approvals
Set what the workflow may read, change and send, where a person approves, and which limits the platform enforces. 2 lessons.
- Module 3
Write tests before trusting the workflow
Build a minimum test set with expected and prohibited actions, including expected escalations. 2 lessons.
- Module 4
Inspect the route as well as the result
Read an action trace against the permitted route and score a run on what it did, not what it said. 2 lessons.
- Module 5
Handle exceptions and stop conditions
Set bounded retries, limits and exits to a named owner, and know when no further action is the right outcome. 2 lessons.
- Module 6
Monitor and change the workflow responsibly
Set owner, sample, measures and version records, and choose release, restricted trial or rollback after a change. 2 lessons.
Built for the person who has to supervise the work, not build the model.
- Operations leads handing bounded work to AI-enabled workflows
- Consultants and coordinators whose assistants prepare and send on their behalf
- Team managers asked whether a workflow is ready to run on its own
- Quality and compliance people asked to sign off on an AI workflow
No programming needed. Comfort using an AI assistant day to day helps.
Written by a quality auditor with 20 years in the industry. The method is an auditor's: define what's allowed, test it, check the record, and decide from the evidence. Every case, person and record in the course is fictional.
One flat price, $129, runs at your pace, and leaves you with an operating brief you can reuse for the next workflow your team wants to switch on.
Straight answers.
What does an expert in my pocket mean?
Can I use what I learn on a real problem?
Can I reuse the work for the next job?
Do I need to code?
Does it connect to my real accounts or AI tools?
How do I earn the certificate?
Who reviews my capstone?
Is this a license or an accreditation?
Is there a time limit on the exam?
How long do I keep access?
Can my company pay?
What if it isn't for me?
Training a team?
Switch it on with evidence.
$129. About 3.5 hours. Your permissions set, your tests written, your traces read, your stop conditions in place.
10% off for 5 or more people.
3-day refund window · Lifetime access to the lessons · Certificate anyone can verify · AI coach included · No subscription