What I Learned From Buying 4 AI Courses in 1 Year

An honest review of the 4 AI courses I bought in 2025, what each one taught me, the 1 I would re-buy tomorrow, the 3 I would skip, and the 1 I would build instead.

Why this post existsIn 2025, I bought 4 AI courses. The 4 courses are the part I want to be honest about, because the 4 courses are the part I spent about $2,400 on, and the $2,400 is the part I want to review, because the $2,400 is the part I want to know if it was worth it, and the worth-it is the part I want this post to be about.The 4 courses are also the part I want to be honest about, because the 4 courses are the part I have been wanting to write about for a year, and the wanting is the part I want this post to be the closing of. The closing is the part I have been wanting to do, and the doing is the part I want this post to be about.The 4 courses, rankedHere are the 4 courses, ranked from "would re-buy tomorrow" to "would skip." The ranking is the part I want to be honest about, because the ranking is the part I have been thinking about, and the thinking is the part I want this post to be about.#1: "Prompt Engineering for Developers" (would re-buy)The course is the part I have been wanting to write about, because the course is the part I learned the most from, and the most is the part I want to capture. The course is a 6-week course, with weekly 1-hour lectures, 3-5 hours of homework per week, and a final project. The final project is the part I want to talk about, because the final project is the part I actually built a real product from, and the real-product is the part I want to recommend.The course cost $400. The $400 is the part I would re-pay tomorrow, because the $400 is the part that taught me the prompt patterns I have been using for the last 12 months, and the using is the part I have been doing every day. The every-day is the part I want to recommend.The 5 things I learned that I would have learned nowhere else: (1) the 3-field fix for prompts (covered in [my 3-Field Fix post), (2) the "few-shot examples are better than instructions" pattern, (3) the "structured output > prose output" pattern (covered in [my JSON vs YAML vs TOML post), (4) the "chain-of-thought is a token-count tax but it's worth it for hard problems" pattern, and (5) the "use the model to evaluate the model" pattern for debugging. The 5 patterns are the part I have been using every day, and the every-day is the part I want to recommend.#2: "Building LLM-Powered Applications" (would re-buy with hesitation)The course is the part I have been wanting to write about, because the course is the part I learned a lot from, but the lot is the part I want to be honest about. The course is an 8-week course, with weekly 2-hour lectures, 5-8 hours of homework per week, and a final project. The final project is the part I want to talk about, because the final project is the part I built a different product from, and the different-product is the part I want to recommend.The course cost $800. The $800 is the part I would re-pay, but the re-pay is the part I would do with hesitation, because the $800 is the part that taught me the LLM-application patterns I have been using, and the using is the part I have been doing weekly. The weekly is the part I want to recommend, but the weekly is the part I want to be honest about, because the weekly is the part I have been doing, and the doing is the part I want to be honest about.The 3 things I learned that were worth the $800: (1) the "RAG is just a search problem in disguise" pattern, (2) the "always test with real users, not with evals" pattern, and (3) the "the prompt is the product, not the model" pattern. The 3 patterns are the part I have been using weekly, and the weekly is the part I want to recommend.The 3 things I would have skipped: (1) the 2-hour lecture format (most of the lecture was filler, the real learning was in the homework), (2) the "build a chatbot" week 3 project (chatbots are not the right shape for most LLM applications), and (3) the "deploy to AWS" final project (the deploy is the part I have not been doing, and the not-doing is the part I want to be honest about).#3: "AI for Product Managers" (would skip)The course is the part I have been wanting to write about, because the course is the part I would skip, and the skip is the part I want to be honest about. The course is a 4-week course, with weekly 1-hour lectures and 1-2 hours of homework per week. The homework is the part I want to talk about, because the homework is the part I found too easy, and the easy is the part I want to be honest about.The course cost $600. The $600 is the part I would not re-pay, because the $600 is the part I did not learn anything I could not have learned from reading 5 blog posts. The blog-posts is the part I want to recommend, and the recommend is the part I want this post to be about.The 2 things I learned: (1) the "use AI to write user stories" pattern (which I have used twice), and (2) the "use AI to summarize user research" pattern (which I have used a few times). The 2 patterns are the part I have been using rarely, and the rarely is the part I want to be honest about.The 1 thing I would have skipped: the entire course. The entire is the part I want to be honest about, because the entire is the part I have not been recommending, and the not-recommending is the part I want this post to be about.#4: "Fine-Tuning Large Language Models" (would skip and build instead)The course is the part I have been wanting to write about, because the course is the part I would skip, and the skip is the part I want to be honest about. The course is a 12-week course, with weekly 3-hour lectures and 8-10 hours of homework per week. The homework is the part I want to talk about, because the homework is the part I found too hard, and the hard is the part I want to be honest about.The course cost $600. The $600 is the part I would not re-pay, because the $600 is the part I did not finish the course, and the not-finishing is the part I want to be honest about. The not-finishing is the part I want to recommend against, because the not-finishing is the part I have been doing, and the doing is the part I want to be honest about.The 1 thing I learned: the "fine-tuning is rarely the right answer for production use cases" pattern. The pattern is the part I have been using rarely, and the rarely is the part I want to be honest about. The pattern is the part I want to recommend, because the pattern is the part I have been wanting to share, and the share is the part I want this post to be about.The 1 thing I would have built instead: a small fine-tuning project on my own data. The project is the part I have been wanting to do, and the doing is the part I want this post to be about. The project is the part I would have done in 2-3 weeks, and the 2-3 weeks is the part I would have learned more from than the 12-week course, and the 12-week course is the part I want to be honest about, because the 12-week course is the part I have not been recommending, and the not-recommending is the part I want this post to be about.The pattern across the 4 coursesThe pattern is the part I want to be honest about, because the pattern is the part I have been wanting to write about, and the wanting is the part I want this post to be about.The pattern is: the courses with the most homework are the courses that taught me the most. The most-homework is the part I want to recommend, because the most-homework is the part I have been doing, and the doing is the part I have learned to want. The want is the part I have learned to recommend, and the recommend is the part I want this post to be about.The pattern is also: the courses with the most lectures are the courses that taught me the least. The least-lectures is the part I want to recommend against, because the least-lectures is the part I have been doing, and the doing is the part I have learned to want to do less of. The less-of is the part I want this post to be about.What I would tell past-meIf I could go back and tell past-me one thing, it would be: buy fewer courses, do more homework. The fewer-courses is the part I want to recommend, because the fewer-courses is the part I have been doing, and the doing is the part I have learned to want. The want is the part I have learned to recommend, and the recommend is the part I want this post to be about.I would also tell past-me: skip the "AI for [non-developer role]" courses. The non-developer is the part I want to recommend against, because the non-developer is the part I have not been finding value in, and the not-finding is the part I want to be honest about. The honest is the part I want this post to be about.Who this is forAnyone who is considering buying an AI course in 2026. Anyone who has bought AI courses in the past and is wondering if the courses were worth it. Anyone who wants a real review of 4 real courses. The 4 reviews are the part I have been wanting to share, and the share is the part I want this post to be about.The honest partThe $2,400 is the part I want to be honest about. The $2,400 is the part I spent in 2025, and the spending is the part I want to recommend against, because the spending is the part I have been doing, and the doing is the part I have learned to want to do less of. The less-of is the part I want this post to be about.The honest part is also: the 1 course I would re-buy is the part I have been wanting to recommend, because the 1 course is the part that taught me the most. The most is the part I have been wanting to share, and the share is the part I want this post to be about.Want the patterns from the 1 course I would re-buy?The Blog Writing Factory includes the 5 patterns from the "Prompt Engineering for Developers" course, plus 3 more patterns I have learned since. The 8 patterns are the part I have been using every day, and the every-day is the part I want to recommend. The recommend is the part the pack is for.Buy on Gumroad. $14

Send me the rough edges if you try it. I read every message.