The February version of this list was written from memory. This one was written from a record.
Since then I’ve kept every video I watch in a notes system — 100 videos, 28 creators, each one filed against the ideas it actually contributed to my work. When I sat down to update this list, I didn’t ask myself who I liked. I checked what I’d come back to.
The record disagreed with my memory in three places. Two of the nine channels I recommended in February, I haven’t watched once since. Three of my four most-watched creators have declined enough that I now recommend them with a caveat. And thirteen creators who weren’t in the original article at all are now in regular rotation.
Here’s the current list, what each channel is for, and the method I use to keep any of it.
Quick reference
| Creator | Best for | Level | Typical length | Videos I’ve logged |
|---|---|---|---|---|
| Dylan Davis | Prompt frameworks and Claude systems | Intermediate | 10–15 min | 17 |
| Ben AI | Skills, routines, marketing automation | Intermediate | 10–20 min | 10 |
| Jack Roberts | Memory systems and agentic builds | Advanced | 15–30 min | 10 |
| Futurepedia | New-tool reviews, non-technical guides | Beginner+ | 10–30 min | 10 |
| ChaseAI | Obsidian + Claude Code, open-source tools | Intermediate | 15–25 min | 5 |
| Jono Catliff | Complete business workflows, SEO | Beginner–Int. | 20–60 min | 5 |
| Kevin Stratvert | Step-by-step tool tutorials | Beginner+ | 10–20 min | 4 |
| Greg Isenberg | Startup thinking, founder interviews | Intermediate | 30–60 min | 4 |
| theMITmonk | AI-era business strategy | Intermediate+ | 15–25 min | 4 |
| Shane Hummus | YouTube growth and AI income | Beginner–Int. | 15–25 min | 4 |
A note on that last column: it counts what I logged, not what I watched. It’s a record of what earned a note, which is the closest honest proxy I have for what earned my attention.
The four I’ve watched most
These four account for nearly half of everything I logged. That number describes the past year, though, and for three of them it no longer describes the present.
Dylan Davis
Prompt frameworks and Claude systems, and the most-watched channel on my list by a distance.
The early work is genuinely excellent. His video on instruction rot — the argument that the best upgrade to your AI setup is deleting half your instructions — is the reason I now audit my own instruction files on a schedule instead of letting them accumulate. The last pass found about 65% duplication between two files I’d assumed were doing completely different jobs.
The caveat: I’ve stopped watching most of what he publishes now. The message has become recycled. It isn’t that the message is wrong — it was so good at the start that the incremental videos feel like an algorithmic cash-grab without the same value per minute. Watch the back catalogue on prompt frameworks and instruction rot. You can skip the recent uploads without missing anything the earlier ones didn’t already say better.
Ben AI
Skills, routines, and marketing automation, built end to end rather than demoed.
His strength is showing a complete pipeline instead of a feature. The de-slopping work — running AI output through a grading pass before you accept it, rather than editing it yourself — changed how I treat a first draft from any model.
The caveat: same trajectory as Dylan, less far along. The recent output leans harder on the format and lighter on new substance. Still worth subscribing to; no longer worth watching everything.
Jack Roberts
Memory systems, design workflows, and agentic builds.
His video on building your first AI agent is the source of the four-part framework I still use to explain what separates an agent from a chatbot — memory, loop, tools, context. It turned into a whole post on this site, which is about as concrete as “this changed how I work” gets.
The caveat: the recent run has drifted toward tool announcements and away from the systems thinking that made the channel worth the time. The memory-system and agent-anatomy videos remain the best things on the channel.
Futurepedia
New-tool reviews and non-technical implementation guides. The one channel of the four I recommend without a caveat.
It works as a discovery engine. When something launches, they test it properly and tell you whether it’s worth your attention, which saves me from evaluating every new tool myself. Their video on answer-engine optimisation — how to get AI tools to recommend your business, rather than how to rank on Google — reframed how I think about who my writing is actually for now.
They’re also the most beginner-safe channel here. If you’re sending one link to someone starting out, send this one.
The rest of the rotation
The ones who teach you how
Kevin Stratvert is still the standard for step-by-step tutorials. Clean, methodical, no padding. His Obsidian beginner tutorial is what finally made me deal with cross-device sync properly, which is a small thing that had been quietly broken in my setup for months. Send him to anyone who says they don’t know where to start.
ChaseAI covers the Obsidian and Claude Code combination better than anyone, plus a steady stream of open-source tools you won’t find through search. If you like the idea of a personal knowledge base that an AI maintains, this is the channel to watch before you build one.
The ones who build in public
Jono Catliff shows complete business workflows end to end — lead generation, SEO, dashboards, web design. His SEO walkthrough is unusually specific about numbers rather than gesturing at results. Best for anyone running a small business who wants the whole pipeline rather than a tactic.
Greg Isenberg is a startup thinker whose interviews are genuinely interesting rather than mutual promotion. Watch for the ideas, not the tutorials. One episode reliably produces several things worth building.
Shane Hummus is blunt about the gap between what AI creators tell you to do and what they actually do to make money. That honesty is the value. Best if you’re weighing YouTube or a personal brand as a business rather than a hobby.
The one who thinks about strategy
theMITmonk covers AI-era business strategy from an actual operating background — former CEO, board member, investor. The argument that the current economy rewards self-reinforcing loops over sequential plans is the most useful strategic frame I picked up this year. Longer on thinking than on tools, which is the point.
Also in the rotation
Worth your time, fewer logged entries: Jeff Su for lunch-break productivity tips, Deya for business fundamentals most AI channels skip entirely, Vicky Zhao and Teacher’s Tech for clear explanations, Greg Kamradt for how frontier engineers actually work, Nicolas Cole on writing, and Aaron Newcomer and ADHDVision on AI and ADHD.
Who I dropped, and why
The Primeagen and Lex Fridman are both out. Neither has contributed a single note since February. I recommended both in the original version of this article and then quietly stopped watching them, which is exactly the kind of thing these lists never admit. They’re good at what they do. What they do stopped being what I needed.
Two others I’ve actively stopped watching, and they share a failure mode worth naming.
Dylan Davis, covered above. ICOR with Tom is the clearer case: the channel is one system, pitched repeatedly. An agent that triages your requests to other agents. It’s a reasonable idea and I understood it the first time.
The pattern is the same in both cases. A creator finds one genuinely good idea, and then monetises the repetition of it. Nothing gets worse, exactly. The value per minute just collapses, because the tenth video is the first video with a new thumbnail.
That’s my actual unsubscribe test now, and it’s more useful than judging quality. Quality is easy to fake for one video. Value per minute is not, and it’s the thing you’re actually spending.
You’ll notice the original list had a “keeps you honest” category and this one doesn’t. Both creators in it were the ones I cut. The skepticism didn’t go away — it moved into the method below, which turns out to be a more reliable place to keep it than in someone else’s upload schedule.
How to actually get more from AI YouTube
Watching more is not the answer. Keeping what you watch is.
Triage with the transcript first. A 40-minute video usually holds about five minutes you need. Grab the transcript, paste it into Claude or ChatGPT, and ask for the key takeaways. Thirty seconds tells you whether to watch, and which section to skip to.
Turn tutorials into checklists immediately. A tutorial is useful while it’s playing and useless the next day when you can’t remember step three. Ask for a step-by-step checklist from the transcript and save it where you’ll actually look. You now have something usable at your desk instead of a video to rewatch.
Synthesise across three or four videos. When you’re deciding something — which tool for a workflow, which approach to a build — watch several and paste all the transcripts into one conversation. Ask what they agree on, where they conflict, and what none of them address. The disagreements are where the real information is.
Fact-check the confident claims. When someone says a tool will replace your entire workflow, paste that section back and ask what the limitations are. Thirty seconds, and it’s most of what the skeptic channels were doing for me.
File it somewhere that compounds. This is the step that makes the other four worth doing. Everything above produces a useful artifact and then loses it in your downloads folder. I keep mine as a set of linked notes in Obsidian, maintained by Claude Code, organised by concept rather than by video — which is how I could write this article from a record instead of from memory. I wrote up how to build that here, and it’s the single highest-leverage thing on this page.
The first four techniques save you time on one video. The fifth is the one that compounds across a year of them.
One honest caveat
This list will be wrong in six months. Three of my top four picked up a caveat in five, and two channels I confidently recommended in February didn’t survive the year.
That’s not a reason to distrust the list. It’s the reason to build your own version of the record behind it. The names on mine expire. The method for knowing which ones still earn your time doesn’t.
No affiliate links, no sponsorships.