Home → Certifications / AI-AAF

Certified Advanced AI Facilitator

Stop changing your own work and start changing how teams work together.

2 daysLive online, capped at 12
8Modules, most of them on your own work
5 + 3Criteria marked, on three bands
For lifeNo renewal fee

Who it is for

You are today Holders of any starting credential
After the class Stop changing your own work and start changing how teams work together.
Prerequisite Any of the three starting credentials (AI-TC, AI-PMPO or AI-PM) admits you.

What you will do

  • Map work that crosses two or more teams
  • Mark every step automate, augment, or hand to a person
  • Design what happens when the AI gets it wrong: the checks, the fallback, the off switch
  • Turn scattered prompts into a shared library the program can run
  • Set the quality bar, and measure whether any of it actually helped

The two days, module by module

Live and instructor-led. You spend most of the time on your own material, because that is what the assessment marks.

  1. 1
    Mapping work across teams, not inside one You map one real flow that crosses at least two teams. 90 min
  2. 2
    The seam: automate, augment, escalate You mark every step in that flow with one of the three calls and write a one-line defense for each that names the property of the step deciding it. 90 min
  3. 3
    Agentic workflow design You write the control specification for one automated step: the decision boundary, the human checkpoint, the fallback, the retry rule and the kill switch, each with a named owning role and a trigger on the kill switch. 120 min
  4. 4
    The shared prompt and pattern library You turn two of your own a starting credential prompts into library entries carrying a name, a version, a tested-against record, a review owner and a retirement trigger. 90 min
  5. 5
    The quality bar You write what the program will and will not accept from AI, in three clauses, and name the single person who decides a contested case and how long they have. 75 min
  6. 6
    Measuring two different things You build a measurement panel with adoption and impact on two separate tracks. 90 min
  7. 7
    The person who will not use it You work one real refusal you brought with you and sort it: a training problem, a tooling problem or a management problem. 75 min
  8. 8
    Building the submission You assemble the six sections of the submission pack, check them against the published rubric with a peer who holds a different starting credential, and run the sanitization pass before you submit. 90 min

How you earn it

Attendance does not earn the credential. Your own work does.

  1. 1 Attend the live class Two days, live online, capped at 12 people.
  2. 2 Submit your own work Within 14 days of class, built on your own material.
  3. 3 An assessor who did not teach you marks it Against the published rubric, back inside 10 working days. It can be failed. One free resubmission.
  4. 4 Your credential, for life A printed Certificate ID and a digital badge. No renewal fee.

What the assessor looks for

Five criteria. Three bands each: Not yet, Pass, Strong. The full rubric opens below.

Not yet Pass Strong
Flow fidelity Your map shows how the work really crosses teams.
Seam calls Every step is marked automate, augment or escalate, with a reason.
Escalation and control design Checkpoints, fallback, retry and a kill switch are named for the automated step.
Measurement separation Adoption metrics and impact metrics are kept apart.
Handover Someone who was not in the room could run it.

What you leave with

One redesigned cross-team workflow
A measurement plan you can defend

What you hold when you pass

Certified Advanced AI Facilitator badge
AI-AAF
Issued by the American School of AI
A digital badge and a printed Certificate ID, verifiable, and yours for life

See the certificate specimen

Certification Standard v1.0

Certified Advanced AI Facilitator AI-AAF

The domains

  1. Mapping work across teams, not inside one
  2. The seam: automate, augment, escalate
  3. Agentic workflow design
  4. The shared prompt and pattern library
  5. The quality bar
  6. Measuring two different things
  7. The person who will not use it
  8. Building the submission

What you submit

After class you submit one pack in six fixed sections: cover block, prompt set, raw output, final version, change log, boundary note. The final version holds the redesigned cross-team flow in four pieces: the handoff record, the seam table with a call and a defense on every step, the control specification for one automated step, and the two-track measurement panel. The prompt set carries two library entries promoted from your a starting credential prompts, and the raw output shows what they returned, unedited, on your own material. The boundary note is your written account of the one step you decided AI must not touch and why, one real refusal sorted with the evidence and the action, and your three-clause quality bar. An assessor who did not teach the class marks the pack against five criteria, each banded Not yet, Pass or Strong. A pass needs Pass or Strong on all five, with no compensation between them. If your employer will not let you bring real material, the supplied practice set is declared on the cover block and marked identically.

How it is marked

Not yet Pass Strong
Flow fidelity

Not yet: the flow sits inside one team, or the second team is a box with no step attached, or no handoff is marked, or the steps are generic stage names with nothing naming what moves between them. Pass: the flow crosses at least two teams, identified by consistent codes from the cover block; every handoff is marked and names what is handed over, who hands it, who receives it, and what the receiving team cannot start without; at least one step carries a real duration or wait. Strong: Pass, and the record names at least one place where the flow differs from how it is documented internally and says which version you are designing against, or it carries a measured wait time at a handoff from a named source.

Seam calls

Not yet: one or more steps carry no call, or the defenses repeat one reason across most steps, or every step carries the same call. Pass: every step carries automate, augment or escalate; each defense names the property of that step that decides the call, such as how reversible it is, who is affected by an error, whether the input is complete, or whether a judgment about a person is involved; at least two different calls appear. Strong: Pass, and at least one change log entry records a call you reversed and what changed your mind, or a step marked escalate names the specific condition that triggers it.

Escalation and control design

Not yet: any of the six required elements is missing (decision boundary, human checkpoint, fallback, retry rule, kill switch, owner), or the kill switch has no statement of who can pull it and what happens to work in flight, or the checkpoint says only that a human reviews the output. Pass: all six present; the decision boundary states what the step may decide and what it must pass on, the checkpoint states what the human checks and what they do when it fails, the fallback states what runs instead, the retry rule states how many attempts and what happens after the last, the kill switch names who can pull it, how, and what happens to work in flight, and each element names an owning role. Strong: Pass, and the design names the signal by which anyone would notice bad output with nobody pulling the switch, or names the failure you expect first from something already observed in that flow.

Measurement separation

Not yet: adoption and impact in one list, or every measure is an adoption measure, or a measure has no source, or time saved appears with no statement of what is subtracted for review. Pass: two separate, labeled tracks; at least three adoption measures with a source, a cadence and an owner; at least two impact measures, each with a baseline that exists now or a stated plan to capture one before the change lands; any time-saved figure states what is netted off for review and rework; at least one measure covers rejected AI output as well as accepted. Strong: Pass, and the plan names a measure that would move the wrong way if the change were working and says why, or names the measure you would put in front of a sponsor first and the number that would make them stop.

Handover

Not yet: a design with no instruction for running it, or the library extract is missing or has no naming convention, version or review owner, or a step depends on knowledge only you have, or the boundary note gives no account a stranger could act on. Pass: the library extract carries a naming convention, a version on every item, a review owner and a review cadence, and the raw output section shows what those entries returned; every step carries an instruction another person in your role could follow; the quality bar says what goes out and what does not; the refusal account is written so a person who was not in the room understands what not to do and why. Strong: Pass, and the library states how an item is retired, or the pack names the first thing that will break when the model changes and what the review owner checks, or the refusal account names a case where the refuser was correct or your own first sort was wrong.

Issuance

You take the two-day live class, then submit one pack built on your own material within 14 days of the last day of class. An assessor who did not teach the class reads it against the published rubric and returns a band and a written reason on each of the five criteria. The credential can be failed. You get one free resubmission, with written feedback, inside 14 days of the mark.

Issued by the American School of AI. Delivered by our authorized training providers. The school develops the standard and has each submission independently assessed. The credential carries a printed Certificate ID and a digital badge.

Anyone can confirm the credential on the Verify a Credential page by entering the Certificate ID printed on the certificate. The credential is for life. There is no renewal fee.
FAQ

Questions, answered

How the Certified Advanced AI Facilitator credential works.

Contact Us
Who is this class for?

Anyone who holds a starting credential from the American School of AI: Scrum Masters and team coaches (AI-TC), Product Owners and backlog owners (AI-PMPO), and project and delivery managers (AI-PM). All of them sit in the same class.

Do I need a starting credential first?

Yes. Any of the three starting credentials admits you.

Who issues the credential, and who teaches the class?

The American School of AI develops the standard, has your submission independently assessed, and issues the credential with a printed Certificate ID. It does not run classes.

How is it assessed?

You submit one pack in six fixed sections, built on your own material, within 14 days of the last day of class. An assessor who did not teach the class, and who never sees your name or employer, reads it against the five published criteria and returns a band and a written reason for each.

What if I do not pass?

Your mark comes back inside 10 working days with written feedback. You get one free resubmission inside 14 days of the mark.

✦Get started

Find a class

The American School of AI does not run classes. Our authorized training providers deliver this one live online over two days, capped at 12 seats.

Find a Class