Greg Isenberg YouTube Breakdown: 5 Lessons for AI Creators

Greg Isenberg's agents and Claude skills explainer reached about 969,000 views, or 8.9 times its nearby normal. His earlier Claude Skills Explained Short landed below its own normal. I measured public view counts for every upload in his Videos and Shorts feeds over the last 2 years, as of November 8, 2026, then read first-minute transcripts and sampled frames from top and bottom videos. The useful question for my next AI tools video is what I can borrow from his teaching craft.

What stood out

Greg teaches agents, coding tools, startup building and personal workflows to about 740,000 subscribers. His main format puts a host and guest beside shared software or a whiteboard. The viewer gets a practical promise: understand the tool well enough to try it.

These were the top 5 qualifying videos from the last 12 months, ranked against each video's own nearby normal:

Video	Approximate views	Times normal
How AI agents & Claude skills work (Clearly Explained)	969,000	8.9
Building AI Agents that actually work (Full Course)	866,000	7.2
How I Use Obsidian + Claude Code to Run My Life	481,000	4.6
Jev is HERE. How to use it	778,000	4.3
Claude Code Clearly Explained (and how to use it)	377,000	4.3

The displayed multiples here follow the narration's rounding. Obsidian ranks above Jev because it beat its own comparison group by more. Named tools arrive with usable lessons, and thumbnails give those lessons a visible shape through diagrams, workflows and products.

1. Name the lesson

Teaching titles went with hits about twice as often: 17 of 46 reached the hit threshold, compared with 33 of 184 other titles. The label includes explained, course, tutorial and demo.

In the April 8 agents and skills explainer, Greg asks what viewers will learn and whether a nontechnical viewer can follow. The guest promises diagrams, and the opening previews a board. The lesson has a visible destination.

Topics and guests overlap across these groups; adding a teaching word alone remains untested. My application would be “Context Windows Clearly Explained,” built around an assistant losing an earlier instruction. Show what information it receives, change the input, and check the behavior.

2. Return with a new angle

Claude titles went with hits about a third of the time: 10 of 32, compared with 40 of 198 other titles, closer to a fifth. In the February 23 Obsidian video, Greg introduces Vin and previews notes beside a terminal. A familiar assistant gets a specific job.

Timing and tool popularity overlap with title choice. August's Claude Code New Features, Explained reached 0.38 times normal, below half. My takeaway is to return with a useful task: explain context, organize a messy repo, then diagnose a lost instruction. Each episode needs a result to demonstrate.

3. Finish the demonstration

Videos from 20 to under 60 minutes went with hits more than twice as often: 47 of 199, compared with 3 of 31 outside that band. Greg's March 17 beginner course names confusing terms, introduces Remy, and previews folders that connect the pieces.

The groups cover different topics, and the outside group combines shorter and longer uploads. The ideal length remains open. I would use the band as a planning budget for a complete learning path: show the initial behavior, open the instruction file, make the change, and rerun the same task. Keep the evidence and cut repetition.

4. Give Shorts their own job

Shorts went with hits somewhat more often: 13 of 42 mature Shorts versus 50 of 230 mature videos. Each feed uses its own nearby normal.

The December 11, 2025 Claude Skills Explained Short reached 0.63 times its normal. The agents and skills explainer the next April reached 8.9 times its own normal. The comparison offers different teaching invitations on a related subject.

The 4 subject-matched pairs include a reused Short and provide anecdotal evidence. Viewer transfer was unmeasured. I'd give a coding Short a narrow question, visible answer and complete payoff, then build a wider lesson that also stands on its own.

5. Introduce through the demo

Within 12 selected openings, teacher introductions went with 7 of 9 videos reaching the hit threshold, compared with 1 of 3 without that opening. In the March 17 course, Remy's introduction gives him a clear job: explain the confusing pieces and their connections, with folders on screen.

These openings were selected for contrasting results, and guests vary. That selection limits what the rates can predict. The craft idea I'd test is practitioner, task, preview: name the repo a guest maintains, show the instruction file they'll explain, and ask them to demonstrate its effect.

How I measured

The snapshot covers November 8, 2024 through November 8, 2026: 235 public Videos-feed uploads and 43 Shorts. Rankings exclude uploads younger than 14 days.

“Normal” is the median of the 20 nearest other qualifying uploads by date, before or after, within the same feed. With that group, the median averages the two middle counts. Divide views by normal to get the multiple; a hit means at least 2 times normal. The explainer's nearby normal was about 109,000 views; the current Videos-feed normal was about 185,000. Each older upload retains its own comparison.

The opening analysis uses first-minute transcripts and sampled frames. Public counts and selected examples support hypotheses to test; they leave topic demand, timing, guests and other differences entangled.

My publishing checklist

1. Name exactly what you'll teach.
2. Give a familiar subject a fresh task.
3. Plan a complete demonstration and cut repetition.
4. Give each Short its own payoff and scorecard.
5. Introduce the teacher through the demonstration.

Which lesson would you try next, and what would your video teach?

Bring your own channel questions to the free 500x Engineer community.