Stop collecting AI tools. Start building systems that hold.
Most people's AI setup is a pile of clever one-offs that quietly rot. This is how to design workflows that still work in six months — with review gates, honest measurement, and a clear rule for when not to automate at all.
Instant access · 30-day refund
Read the free version first. The core idea in this course is covered in My AI Workflow — And the Three Places It Still Fails — free, no email required. Buy the course only if you want the full system, the templates and the walkthroughs.
This is for you if
- You've got a dozen AI subscriptions and can't point at an hour they saved.
- You automate things and then quietly stop trusting the output.
- You run a small business or a solo operation and you're the bottleneck.
- You want the failure modes explained, not hand-waved.
Skip it if
- You want to build and sell AI products — this is about operating, not shipping software.
- You're expecting prompt lists to copy and paste.
- You need an enterprise governance framework.
- You believe the demos.
What you'll have at the end
Not "an understanding of". Actual artefacts you can point at.
- A mapped inventory of your own work, with every task scored on whether automating it is actually worth it.
- Three live workflows built end to end, with review gates that catch the failure modes before your customer does.
- A prompt architecture you can maintain — versioned, testable, and readable by future-you.
- A measurement sheet that shows real hours saved rather than hours you assume you saved.
- A written kill-list: the tasks you've decided to keep doing by hand, and why.
Every module, every lesson
Click any module to expand it. Nothing is hidden before you buy.
- Mapping your week in 90 minutes
- The four-factor score: frequency, cost of error, verifiability, stability
- Why high-frequency + low-verifiability is the trap everyone falls into
- Worksheet: your scored task inventory
- Structure that survives model changes
- Instructions vs. context vs. examples — what belongs where
- Versioning prompts like code, without pretending you're a developer
- Building a small test set before you trust anything
- Designing the check that catches the 5% failure
- Sampling rates: how much output you actually need to read
- The three tasks I will never run unreviewed, and why
- End-to-end build, screen recording, no cuts
- Where it breaks and how I caught it
- Handling sources it can't access
- The two-pass method that keeps your voice
- Why 'make it better' produces worse output
- Editing gates that stop generic writing reaching the page
- Triage, tagging and routing
- The escalation rule for anything touching money
- Logging so you can audit it later
- The measurement sheet (and honest accounting for review time)
- The quarterly audit that catches rot
- Deciding to kill an automation without sunk-cost guilt
From students who finished it
“I've read a hundred posts about AI workflows. This is the first one that was honest about where it breaks.”
“The four-factor score killed six automations I'd been maintaining out of pride. That module alone was worth it.”
“Genuinely useful and refreshingly unhyped. It assumes you're an adult with a business to run.”
Arie — nine years, forty countries, one laptop
I've been running an online business from wherever I happened to be since 2017. I teach this because I got most of it wrong first, expensively, and writing it down is the only way I've found to make that cost worth something. More about me →
Questions people actually ask
AI Systems for Solo Operators
Automation that still works in six months. One payment, lifetime access, and 30 days to decide it was worth it.
Enrol now — $297