Most free AI courses teach you about AI. This one changes how one real task in your own job gets done, and hands you the evidence to prove it.
Fair question. There are genuinely good free AI courses already, and we had no interest in building a ninth version of the same one.
We took apart the eight largest free AI courses. Every one teaches the same six things: what AI is, how to prompt it, a tour of use cases, ethics and bias, the future of work, and how to check the output. That material is genuinely useful, and it is genuinely everywhere — Google, Microsoft, IBM, OpenAI and Anthropic will all teach it to you for nothing.
What none of them do is touch your actual job. Every exercise is invented, the assessment is multiple choice, and you finish holding nothing you did not start with. The free courses aimed at Africa are built for job seekers in their twenties and ask for 120 to 180 hours. Nobody is serving people who already have the job.
So Foundations skips all of it. Twenty sessions, four hours, spent entirely on your task, your sector, your language, your currency, your country's law and your employer's permission.
Most people here have already tried these tools. Very few use them every day, and almost nobody has permission, a cost model, or any idea what the law requires of them. That gap — between having heard of AI and actually working with it — is the whole point of this course.
AI diffusion in Kenya as a share of working-age population, against 27.5% across the global north
Microsoft AI Diffusion Report, Q1 2026
African workers using AI daily, despite 64% having used it at least once in the past year
Regional adoption surveys, 2025-26
Points of accuracy the strongest models lose moving from English to African languages; open models lose up to 37
AfroBench, McGill & Masakhane
Median token premium on African languages, rising to 7.36× for Amharic — you pay more for the same sentence
Tokenizer analysis across major models
Professionals using AI tools their employer has not approved; 58% have pasted sensitive data into one
Shadow AI workplace studies, 2025-26
Organisations with any formal AI policy at all, leaving most people guessing at what is permitted
Enterprise AI governance surveys, 2026
5 lessons, 20 sessions, about 4 hours in total. Every session ends with something in your hands.
Before you touch a single AI tool
Everyone else starts with the technology. We start with your timesheet, because you cannot tell whether AI helped if you never measured what things cost you beforehand.
Open lesson 01Break a typical working week into tasks and put an honest number of hours against each one. Most people are surprised by at least one row.
The short, non-technical explanation of what these systems are doing, followed by the failure modes that matter at work: fabrication, stale knowledge, weak arithmetic, and no idea what it does not know.
A short diagnostic across your data, your tools, your permissions and your confidence. Scored, so you can repeat it in session 5.4 and see the movement.
Pick your track — SACCO and cooperative finance, mobile money reconciliation, county government, NGO and donor reporting, agri-cooperative, MSME and informal trade, or corporate knowledge work — then choose the one task you will carry through the remaining sixteen sessions. Selection criteria: frequent, text-heavy, low-risk and genuinely yours.
On your task, not on an exercise
Prompting is the one thing every course covers. The difference here is that you are prompting against your own documents and your own quality bar from the first minute.
Open lesson 02A five-part structure — context, role, input, standard, format — that you can reuse for anything. Applied immediately to your anchor task.
Feed the model your real material: a report, a policy, a spreadsheet, a set of meeting notes. Learn what to strip out before you do, and why.
Every course warns you about hallucination in the abstract. This one shows you the local failure surface: invented Kenyan and Nigerian case citations, out-of-date KRA, FIRS and SARS rules, fabricated national statistics, and currency and date formats quietly rendered American. Build a checking routine short enough to survive a deadline.
Run your anchor task end to end with AI for the first time. Time it. Compare against the figure from session 1.1.
Language, cost, connectivity and where your data goes
No major free course covers any of this. It is also the lesson that decides whether AI is usable in your actual working life, or only in a demo.
Open lesson 03Run your anchor task in English, then in Swahili, Sheng, Pidgin, Yoruba or Amharic, and score both. Benchmarks put the drop at about 19 points for the strongest models and up to 37 for open ones, with only around 2% of African languages supported at all. Then learn the mitigations: translate-work-translate, a term glossary, and knowing when to simply stay in English.
A subscription runs about 5% of an average monthly salary, and your language carries a premium on top — roughly 1.5× for Swahili, 1.85× for Yoruba, over 7× for Amharic, because the same sentence costs more tokens. Price your own usage across free tiers, the cheaper regional plans and pay-as-you-go, and work out what you genuinely need to pay for.
Fixed broadband is under 1% of connections and most people work from a handset. Batching prompts, drafting offline, reusable prompt blocks and keeping sessions short — all demonstrated under 500 kbps, which is also the budget this course itself is built to.
Pasting client data into a foreign model is a cross-border transfer of personal data, and it needs a lawful basis. Work through Kenya's Data Protection Act sections 48 and 50, Nigeria's NDPA section 41 and South Africa's POPIA section 72 against your own task, then classify what you handle into what may go, what may go redacted, and what never leaves.
The part every other course leaves out
Knowing how to use AI is useless if using it gets you into trouble. Most people in African workplaces are already using these tools quietly, without permission and without disclosure. This lesson fixes that.
Open lesson 04Around 80% of professionals use AI tools their employer has not approved, 58% have pasted sensitive data into one, and senior people offend most. Roughly half say they would carry on even if it were banned outright. What that exposes you to personally, and why the quiet route is the risky one.
Find your organisation's AI or acceptable-use policy and work out what it actually permits. Fewer than a third of organisations have one, so this session also covers what to assume in the vacuum, and how to work compliantly when the tools themselves are blocked.
When disclosure is required, when it is merely wise, and how to word it to a manager, a client or a regulator without undermining your own work.
Write a single page for the person who actually signs off: the task, the change, the hours saved, the data classification from session 3.4, and the risks with how you have handled each. Peer-reviewed against one question — would a real manager say yes to this?
What changed, and what it was worth
A certificate says you watched some videos. Evidence says you changed how work gets done. You finish this course with the second kind.
Open lesson 05Re-time your anchor task against the baseline from session 1.1 and convert the difference into hours a month and money a year at your own charge-out rate. Includes the discipline of not flattering your own numbers, and the arithmetic for when the honest answer is zero.
Write down what did not work, where the model let you down, and what you stopped trying. Nobody else will tell you where these tools fail, and it is the part that makes the rest of your dossier credible.
Assemble the audit, the prompt, the red-line list, the memo, the measurements and the failure log into one document, then exchange it with someone in your sector track. Self-paced courses finish at around one in eight; reviewed cohorts finish at four in five, and this session is why.
Re-score your readiness, commit publicly to a thirty-day habit with a fixed re-measurement date, and pick your next two tasks. Only 17% of African AI users are daily users; this session is about joining them.
Six artefacts, not a completion badge. Together they make up your AI Work Dossier.
An honest picture of where your hours actually go, which most people have never written down.
Not a template. A brief tuned to your documents, your standards and your output format.
The specific things you will never paste into a model, derived from the law that applies to you.
The task, the change, the time saved and the risks handled — ready to send to your manager.
Real timings on your own task, converted into hours a month and money a year.
All of the above in one short document you own, plus a failure log that makes it credible.
The twenty sessions are free and always will be. If you want an independently assessed, verifiable credential to put in front of an employer, you can sit the Foundations assessment at the end. It marks your dossier and your applied work, not your video-watching.
Nothing in the course is held back if you choose not to. Assessment dates and fees are confirmed at the end of each intake.
Ask about assessmentFoundations is step one of six. You are free to stop after it — plenty of people do, and they still leave better off than they arrived.
The ones people ask before they sign up.
Yes. All twenty sessions, at no cost, and we never ask for payment details. The paid programmes exist if you want to go further afterwards, but nothing in Foundations is withheld behind a fee.
None. No coding, no maths, no prior AI experience. If you write documents, handle data, answer email or prepare reports for a living, the course is aimed at you.
Then lesson four is the most valuable part of the course for you. It covers reading the policy, understanding what it actually prohibits, and making a documented case for sanctioned use rather than quiet use.
About four hours in total across 20 sessions. Most sessions run ten to sixteen minutes. You can take one lesson a week or work through the whole thing in an afternoon.
The course is built to be completed on a phone, and session 3.3 is specifically about working well on a mid-range handset and a patchy connection.
That is a legitimate finding and the course treats it as one. Session 5.2 is a failure log. Knowing where these tools do not help is worth as much as knowing where they do, and it is more than most courses will ever tell you.
Because it is the single thing that makes people finish. Self-paced online courses complete at roughly one in eight; cohorts with reviewed work complete at around four in five. Sessions run on your own schedule, but each intake has fixed dates and you exchange your dossier with someone in your sector track at the end.
No. Everything in the course can be done on free tiers. Session 3.2 exists so that if you do decide to pay, you know exactly what for, what it costs in shillings, and what the cheaper regional plans do and do not include.
The same twenty sessions, run as a cohort inside your organisation, with your own tasks and your own policy.