01
Understand before automating
- Day 01AI, automation, and agents
- Day 02Draw a workflow before building
- Day 03Seven useful AI task patterns
- Day 04JSON, APIs, webhooks, and OAuth
- Day 05Instructions and structured outputs
30 days · built for Kenya · beginner friendly
Build AI that does useful work.
Go from “AI ni nini?” to a tested business agent you can demonstrate. You will map workflows, connect APIs, build with n8n and code, protect customer data, test failures and package a believable Kenyan client proposal.
No hype and no assumed tech background. English comes first; Kiswahili, Sheng, +254 and low-connectivity cases appear in the test pack.
Daily target1 focused lesson
About 60–75 minutes. Your bar remembers the day in Africa/Nairobi time.
Read + watch20 minutes
Plain-language notes and a verified visual companion.
Build + break30–45 minutes
Follow the lab, then run normal, edge and failure tests.
Low-energy dayTired Mode
A diagram, five essentials, three recall prompts and one-sentence wrap.
Nothing important hidden
Thirty days, from first principles to a client-ready system.
Six chapters keep the pace sensible. Every day has a Kenyan example, setup or build steps, a concrete test, a recap and exactly what to memorise.
02
Build visible no-code workflows
- Day 06Build your first n8n workflow
- Day 07Receive and clean webhook data
- Day 08Classify enquiries with AI
- Day 09Reliable actions and approvals
- Day 10Ship the no-code triage lab
03
Engineer bounded agents
- Day 11Inside an AI agent
- Day 12Call the Responses API
- Day 13Function calling safely
- Day 14Build with the Agents SDK
- Day 15Orchestrate multiple specialists
04
Ground, measure and govern
- Day 16Ground answers with RAG
- Day 17Use memory with restraint
- Day 18Build an evaluation set
- Day 19Trace cost, speed, and failures
- Day 20Guardrails and human review
05
Keep Kenyan systems safe
- Day 21Kenya data protection in practice
- Day 22Secure tools, secrets, and access
- Day 23Learn with Daraja sandbox
- Day 24Design consent-aware messaging
- Day 25Deploy, monitor, and recover
06
Turn proof into a business case
- Day 26Discover the right workflow
- Day 27Model value without hype
- Day 28Scope and sell responsibly
- Day 29Build the Karibu Desk capstone
- Day 30Prove, present, and improve
Proof beats promises
Build systems a Kenyan business owner can inspect.
All records are fictional. Live customer, health, school, SACCO and payment data stay out of the learning environment.
Enquiry triage that drafts, never guesses.
A form or webhook validates a +254 number, classifies English/Kiswahili intent, writes one CRM row and prepares a grounded reply.
- Duplicate-event protection
- Missing-field and injection tests
- Human handoff for uncertainty
Laundry status without WhatsApp chaos.
Route pickup requests, calculate a deterministic quote, send reminders and keep an auditable status history—even when a webhook retries.
- Africa/Nairobi scheduling
- Offline and retry path
- Consent and opt-out handling
One bounded agent, ready for a real demo.
Combine grounded knowledge, one typed business tool, approvals, CRM logging, evaluation, observability, cost estimates and a proposal.
- Daraja sandbox only
- Golden-set regression tests
- Rollback and client handover
Day 0 is part of the work
Set up a safe practice desk before touching an agent.
The full course gives Linux, macOS and Windows-friendly paths. No secret belongs in a prompt, screenshot, public repository or browser bundle.
Create a fictional business
Use Karibu Bites, Mwangaza Salon or Safisha Laundry—never real customer records.
Choose a build route
Start with n8n Cloud or a local Docker sandbox; Make and Zapier are compared, not forced.
Store credentials correctly
Use credential vaults and environment variables with capped API spend and separate test accounts.
Prepare synthetic fixtures
Include valid, missing, duplicate, malicious, Kiswahili, misspelled and timeout cases.
Prove the setup
Run a health check, send one fake webhook, inspect the log and remove the test record.
Pause, copy, then try alone
Verified visual companions—not a substitute for testing.
Titles and channels were checked on 16 August 2026. Interfaces change, so the written checklist and official docs remain the source of truth.
What’s an AI Agent?OpenAI · mental modelBeginner course: introduction to automationn8n · workflow basics
APIs and webhooksn8n · connect systemsBuild Hour: Responses APIOpenAI · current code route
Build Hour: Agents SDKOpenAI · tools and orchestrationBuild an agent with human approvalMake · approval boundaries
Measuring agents with evaluationsOpenAI · evidence over vibesTop ways to attack LLM applicationsIBM Technology · OWASP overview
Monthly learning access
One month. Thirty serious steps forward.
The course price covers the KODE Ń VIBE learning material and browser study tools. Any OpenAI usage, mobile data, external automation plan, hosting or client infrastructure is separate; free tiers or local sandboxes are used where practical.
Researched, dated, honest
Built from your notes, then checked against current primary sources.
Four supplied PDFs were deduplicated into one learning sequence. Unsupported marketing numbers and UAE pricing were removed rather than converted into fake Kenyan evidence.
Current AI practice
New code uses the OpenAI Responses API and Agents SDK. Retiring Assistants and legacy Evals material is identified as history, not taught as the future.
OpenAI deprecations ↗Kenyan data duties
Consent, purpose limitation, minimisation, retention, automated decisions and cloud transfers are introduced through Kenya Law and ODPC resources.
Data Protection Act ↗Payments stay bounded
Daraja appears only through sandbox practice, HTTPS callbacks, server verification, replay safety and human confirmation—never browser secrets or blind “success” messages.
Safaricom Daraja ↗Before you pay
Straight answers.
If a detail affects cost, safety or access, it should not be hidden in small print.
Can a complete beginner follow this?
Yes. Day 0 covers accounts, safe sample data and tool setup. The course defines each acronym in plain English before using it, then gives an optional code route for learners who want deeper control.
Is the KSh 5,499 payment an automatic subscription?
No. It grants 30 days of access and does not renew automatically. If you want another month, you make another deliberate payment.
Are OpenAI, n8n, Make, Zapier, hosting and data bundles included?
No. Those are third-party costs. The course starts with local sandboxes and free allowances where available, teaches spend caps, and tells you before a lab could incur usage charges.
Will I connect a real M-PESA till or customer list?
No. Learning uses Safaricom’s Daraja sandbox and fictional records. Production credentials, financial data and real customer details require a separate security, legal and operational review.
What happens on a tired day?
Open Tired Mode from the universal bar. It gives the current lesson as a diagram, five key ideas, a Kenyan example, three recall questions and one sentence to remember. The quick read can hit your reading target, but practical labs remain visible.