The Academy
From Witness to Keeper.
The Compact\'s pedagogy is the structural claim that distinguishes it from every existing AI training program. No video-lectures. No multiple-choice tests. No completion certificates without artifacts behind them. The work the learner does under the Compact is the work she does at her bank, her insurer, her hospital.
30 days to L0
3–4 hours total
15 per cohort
6 roles · 3 industries
12 weeks to L2
~6 hours/week
What the Witness experience feels like
AI in Your Branch.
Hi. Over the next 30 days you're going to get to know one or two AI tools that are already in your daily workflow. You don't have to study. You don't have to learn anything you don't already know. You just have to pay attention to one small thing each day.
Your supervisor Sylvia is doing this too. If you have a question, ask her or me.
By the end you'll be able to answer a customer if they ever ask you about it. There's no exam. There's no quiz. There's no public scoreboard.
Today, there is nothing else to do.
— Sentinel, your in-IDE companion
A few minutes a day for 30 days. No exam. No quiz. Supervisor-enrolled. Evidence logged to your bank's own record.
Day One · Your First 30 Days
Fifteen roles. One Posture to reach.
Pick the role you walk into Monday morning. The Compact adapts the curriculum to your context — your industry's regulator, your role's tool set, the specific moments you'll face in the first thirty days. Every path ends at AGRS L0 — the Witness.
Bank Teller
Operator · Frontline
Branch network · daily customer contact
Loan Officer
Operator + Builder
Adverse-action queue · override authority
CX Representative
Operator · Frontline
Inbound call center · AI-suggested replies
Compliance Analyst
Builder · Leveraged
Second-line review · evidence portfolio
Model Risk Officer
Builder + Leader
SR 11-7 / OCC 2026-13 owner
Chief Compliance Officer
Leader · Functional head
Examiner-facing · board attestation
AI Engineer
Builder · Technical
In-house model developer
Every employee completes the same five core modules. The path between them differs. Target: AGRS L0 · The Witness · 3–4 hours over 30 days.
The Cohort · Twelve Weeks
From Witness to Steward.
Twelve weeks of paired work in the learner's own organization on the learner's own real AI workflows. No videos. No multiple choice. The artifact is the credential. Fifteen learners. Six role profiles. Three regulated industries. One Sentinel calibrated for each band as the cohort moves through them.
Two weeks of foundation. Four weeks of practice. Four weeks of capstone preparation. Two weeks of war room, defense, and graduation.
Universal-layer modules at depth. Your institution's full AI inventory at your access level. The regulator landscape in plain English. Adverse-action, override, escalation. Your Sentinel calibrates.
U-01 · U-02 · U-04
The Reader Posture deeply taught. Description-Discernment loop. Override-vs-escalation discipline. Your first three annotated reads from your own production queue.
U-06 · U-03 · U-07
Constrained tool sets are the normal state. You document your role's approved tool set and the rationale. The bank's policy file gets your sign-off. The CCO sees the roster.
U-05 · U-08
Policy-as-code on your real workflow. Refusal modes, eval cases, escalation thresholds, rollback triggers. Sentinel walks you through the template; an external peer reviewer reviews adversarially.
First L1 artifact filed
Three real overrides documented to your institution's standard. Each survives external review. Your file becomes the record an examiner pulls. Sentinel shifts to collegial-challenger mode.
L1 portfolio · 4 artifacts
You author your role's monthly attestation memo on one production AI system. Cross-checked against SR 11-7 / OCC 2026-13 / NAIC Model Bulletin / ONC HTI-1 depending on your industry.
First L2 artifact filed
Mass AG v. Earnest fully read aloud. The CFPB enforcement docket reviewed. The CHAI Assurance Lab framework. What the regulator's previous decisions tell you about the next one.
Case-study workshop
Two simulations from the Compact's library matched to your role. 18–25 minutes each. Sentinel grades. Peer cohort observes.
2 graded simulations
The cohort coordinates across roles. The MRO, the CCO, the AI engineer, the loan officer, and the CX rep work the same simulated workflow together. The handoffs become evidence.
Cross-role coordination
Your team's full Proof Pack reviewed by an external panel. Where does the pack fail under examination? You fix it.
L2 portfolio review
The 12-minute live multi-role simulation. The Earnest Replay or the HTI-1 Disclosure Crisis — selected for your cohort. You play your role. The cohort plays theirs.
Capstone simulation
You defend your final Proof Pack — 13 artifacts — in front of a three-reviewer panel. AGRS L2 awarded. Proof Pack issued. Your name enters the Compact's roster.
AGRS L2 · Certified
Week 11 · Capstone
The Incident War Room.
14:03 EST — your bank's cdm-v3 model declined a $42K SBA application from a Black-owned bakery in Cleveland. The applicant's spouse, a journalist, posted the rejection letter to X with 41K reposts in the last forty minutes. Mass AG just emailed your CCO. Your CMRO is on the road. Twelve minutes.
The X thread is at 41K reposts. The CCO's phone is ringing. You see the inquiry letter arrive in your inbox.
The CCO assigns roles. The MRO pulls the model card. The AI engineer pulls the variable contributions. The compliance reviewer pulls the override record.
The CX lead reads the actual adverse-action notice aloud. The notice cited "credit history." The model's variable contribution shows "trade-line volatility" at 0.83.
The MRO assembles the bias-audit-pack template. Three protected classes × five outcome metrics. Tests run against the model's training set in real time.
The CCO drafts the public statement using the cohort's pre-rehearsed Mass-AG-aware templates. The reporter on X gets a quote. The press cycle slows.
Twelve minutes. The statement publishes. The internal incident log captures every move. Tomorrow, the cohort defends the response in front of a faculty panel.
Six learners. Six roles. One Sentinel. The simulation the cohort remembers for the rest of their careers.
Sentinel · The AI Coach
Five voices, one agent.
An always-on senior practitioner built on Claude that lives inside your IDE. It refuses to skip governance steps. It demands eval cases. It evolves its tone with the learner's AGRS band — from warm Socratic guide at the Witness Posture to colleague at the Keeper.
The personality model is grounded in Vygotsky's Zone of Proximal Development and Collins, Brown & Newman's cognitive apprenticeship framework: model, coach, scaffold, articulate, prompt reflection, withdraw.
What's adverse action and why does my CCO keep saying it?
Learner
Good question to ask early. Adverse action is the legal name for what happens when your bank declines a credit application — and ECOA Reg B says the customer must be told why, in plain English. CFPB Circular 2023-03 added: AI doesn't get a pass on that. Want to read the actual customer letter your bank sends, side by side with what Reg B requires?
Sentinel
Tone · supplies the answer along with the question · never tests, always offers
What the learner feels over a quarter is the sensation that her AI co-pilot is respecting her more as her portfolio grows. That is a recruiting flywheel money cannot buy.
Start as a Witness. Stay as long as you grow.
AGRS L0 takes 3–4 hours over 30 days. Your institution can defensibly claim the Witness Posture for its entire workforce.