#1
Workers rank quality control of AI output as the top skill, ahead of critical thinking.
Microsoft Work Trend Index 2026Shapes our design:School 4, Evals and Verification.
Build your first AI helper
AI coaches + real people
Learn at your pace
Founding cohort waitlist open
Learn with joy. Build with AI.
Warm teachers, real work and AI coaches that help you think, not think for you. Join the founding cohort of a school you will look forward to.
- 9Schools
- 46Lessons
- 5Ladder levels
This is what learning feels like here
Illustration
Which one sounds like you?
Be the classmate who builds the cool thing.
Start Literate. Learn to use AI without losing your own thinking, then make something your friends want to try.
Your first win
A failure log of an AI tool you tried, with what it got wrong.
Lessons on this path
The curriculum
Nine schools. Forty-six lessons.
Every lesson ends in something you made and can show a friend.
Model Literacy
How models work, fail and leak, so you trust them exactly as much as they deserve.
Tools and Context
Connect models to the world: functions, protocols, retrieval and memory.
Agent Orchestration
Loops, handoffs, gates and permissions: how many agents do real work safely.
Evals and Verification
The skill employers rank first: knowing whether the output is good.
Domain Judgment
Frame problems worth solving and price the cost of being wrong.
Taste and Expression
Critique, narrative and ethics: the human side of making things people trust.
Science and Method
Design experiments that can fail, pre-register, replicate and forecast honestly.
Governance and Civic
Policy, security, labour and public-sector deployment under oversight.
Creative Studio
Make film, music and games with AI as your crew, while you keep the taste, the vision and the credit.
Film, music and games
The ladder
Five levels. Each one proven.
Start here
You will complete
Proof
A documented ten-task AI-assisted workflow with an error log.
You will complete
Proof
A tool-using assistant that passes a 20-case eval set at a threshold you set in advance.
You will complete
Proof
An agent serving real users for 30 days, with traces and an incident record.
You will complete
Proof
A gated multi-agent system that beats a baseline on a pre-registered eval, reproduced by a peer. You mentor one Builder.
You will complete
Proof
A standard adopted by two outside teams, plus an audit of another team's system that finds and fixes real faults.
Steward
AI coach
Human Dean signs off
Faculty
Meet your AI coaches
Every message says it is AI. Real people, our Deans, sign off on every grade.
Professor Ndovu
Scripted sample, not a live chat
- You
Why does my agent keep repeating the same tool call?
- Tool call: Retrieved lesson 3.1, section on loop termination
- AI coach · reviewed by a Dean
Good question to ask. What does your loop check after each tool result, and what would tell it to stop?
Guides you with questions and a hint ladder, one per course.
ToolsCourse retrievalYour learner profile (read)Guardrail Never gives final answers on graded work.
Agents teach. Humans decide.
Dean
Owns the curriculum, handles grading appeals and signs off every grade and credential.
Steward
Handles safety escalations, child protection and data-subject requests.
Faculty lead
Owns the quality of one course and reviews a weekly sample of tutor transcripts.
Do you carry that judgment? Teach with us → Read the free playbook →
School 9 · Creative Studio
Direct the film. Compose the track. Design the game.
Make film, music and games with AI as your crew, while you keep the taste, the vision and the credit.
Also in School 9
AI is your crew. You are the director — and you keep the credit.
Community
Learn together, not ranked.
Study pods of friends + one AI peer
Choose a house to see its shared goal and the principle behind it.
- Shared goals that a whole house works toward.
- Personal bests, never a public league table.
- One AI peer per study pod.
Proof ledger
No single best student. A proof ledger.
Nobody is ranked against a classmate. Each learner builds a ledger of measured outputs, and a profile across them shows where the work is strong and where it is not yet.
1Eval delta2Shipped impact3Catch rate4Calibration5Orchestration yield6Reproducibility7Judgment under ambiguity
Example profile — illustration
Tap a metric to see what it measures and to highlight its axis.
Open lab
The lab is open.
Learners work on real artefacts, not exercises made up for a course. Here are three you can run today.
45
parameters
Scenario model
Five conditional scenarios across 45 named parameters. Run it yourself.
scenario_model.pyOpen Scenario model (opens in a new tab)5
scenarios
Model results
Every scenario and sensitivity as data, with the evidence boundary written into the file.
model-results.jsonOpen Model results (opens in a new tab)363
source URLs
Methodology
What the model can say, and what it cannot. The inventory holds 382 dated claim lines and 363 distinct URLs, and it is an inventory, not 382 established facts.
methodology.mdOpen Methodology (opens in a new tab)
python scenario_model.py“These are conditional calculations, with no scenario probabilities or fitted causal coefficients.”
For organisations
Teams learn AI on their own tasks.
Enterprise is different from open enrolment. We start inside your organisation: we map what each role really does, then teach named employees how to use AI on their own tasks, one role at a time.
- Executive & Board
- Clinicians
- Operations & Admin
- Customer Support & WhatsApp
- Finance, Credit & Underwriting
- HR & Learning
- Engineering, IT & Data
- Risk, Legal & Compliance
- Sales, Marketing & Comms
- Public-Sector Officers
- Diagnose
- Map tasks to lessons
- Lane cohorts
- Field work
- Before and after
- Lane Champions
- Scale
Evidence
What the research says — and what we built because of it.
Some of these results are early or limited. We say so, and we design around them.
Read this first
-17%
Students who practised with a plain AI tutor scored 17% lower on later unaided exams. A tutor built around hints removed the harm.
Bastani et al., PNAS 2025Shapes our design: Unaided first. Hint ladder.
n=194
A purpose-built, scaffolded AI tutor beat an in-class active-learning lesson in a Harvard physics trial. One course, so we treat it as promising, not proven.
Kestin et al., Scientific Reports 2025Shapes our design:Design the tutor; the model alone is not the ingredient.
15%
In a 2026 study, an AI math tutor showed no gain over practice alone, and only 15% of students with access used it.
Hechinger Report, 2026Shapes our design:Tutors alone do not drive engagement. Cohorts and humans do.
28%
Job postings that ask for AI skills carry a 28% wage premium.
Lightcast, 2025Shapes our design:Why institutions pay for the Ladder.
33%
Only a third of workers received employer-provided AI training in the past six months.
The Conference Board, 2026Shapes our design:The gap institutions have today.
The market
Why now.
Planning bands, not forecasts
AI-powered corporate training
2026$7.49B
2031$18.19B
USD billions
Source: Mordor Intelligence
AI certification courses
2026$4.72B
2031$12.36B
USD billions
Source: Mordor Intelligence
Market estimates vary by definition. These are two analyst series, shown as planning bands, not forecasts we endorse.
The big labs now give AI literacy away free. What is scarce is verified judgment, supervised practice and sector-specific proof.
Founding cohort waitlist open
Be in the founding cohort.
Tell us which path sounds like you. We will write with dates and a seat. Joining is free, and you can change your path any time.
- Free to join, no payment details
- We write to you with dates and your seat
- Change your path whenever you like
Prefer to talk? Message us on WhatsApp.
Founding cohort
Save your place.
Tell us a little about you. Dates and seats are set soon, and we write when they are.
Questions
Honest answers.
We are opening a founding cohort. Join the waitlist and we will write to you with dates and a seat. Nothing is charged to join.
AI faculty teach and coach, and every message says so. Named humans, our Deans, sign off on every grade and credential. A human mentor is one tap away.
That is the risk we designed against. You work unaided first, faculty give hints rather than answers on graded work, and assessment is always unaided.
No. The Literate level and School 6, Taste and Expression, need none. The Builder and Orchestrator levels do involve building, and we teach it.
Yes. School 9, Creative Studio, has six lessons: directing film, composing music, designing games, worldbuilding, rights and credit, and releasing a work. You direct and decide; AI is your crew. You finish with something real that an audience has seen or heard.
A team of learners with shared progress toward a goal. You see your own personal best, and no public league table of individuals.
Free courses teach tools well. We teach evals, domain judgment and experimental method, and we certify proof you can show a peer, not hours watched.
Through six cohort packages, scoped in writing. Start with a Discovery conversation and we will map a package to your people.
We collect only what a waitlist needs. Learner profiles are visible and correctable by you, and children's data requires guardian consent under Kenya's Data Protection Act.
Both, in different ways. Individuals pick a path and join the founding cohort. Companies and institutions start with a conversation: we map what each role does, then teach named employees on their own tasks.
A Discovery Diagnostic maps your workflows. Each role gets a lane: a short cohort that practises on that role's real tasks, unaided first, with a human checkpoint. We then measure the same tasks before and after, and a Dean signs the proof report.
Images are AI-generated illustrations of the learning we are building. They are not photos of real learners.