A Thousand Agents and a Pulsing Blob: Eric Morrison Roasts the Agentic AI Hustle

Meme is where we look at the funniest takes on AI culture, and what the joke gets right. The first one is a roast of a social-media genre most people working with AI will recognise: the solo founder whose voice assistant reads out a perfect business report.

In a July 2026 video, Eric Morrison takes aim at the creators who present themselves as independent “agentic AI hustlers” (Morrison, 2026). The video is embedded below; the timestamps jump to the matching moment. This post is about the trend and its claims. The clips Morrison reacts to are left unnamed here.

The advice, turned up to eleven

The video opens with the advice in its purest form. Your agents should work around the clock, especially while you sleep. You should not prompt them; you should build loops. Send them on tasks that run for weeks, and do not ask how many tokens that burns. Have them write the content, answer the customers, design the product and set the strategy, and give each agent its own swarm of sub-agents (watch from 0:00).

Then comes a real clip: SoftBank chief executive Masayoshi Son predicting that a single person will direct the work of a thousand agents. Morrison runs with it. If companies cannot find a return on AI, they must simply be running too few agents, so obviously the answer is a thousand, then more than a thousand (watch from 0:40; 1:00).

Why bother roasting it

Morrison is clear that agents are an exciting technology that empowers experts. The complaint is that the same tools are also, in a phrase from the video, “turbocharging goobers and cringe lords” (watch from 1:40). The reason for caring is not that the content is silly. Morrison argues that viewers take it seriously, and come away believing they must hand everything to agents or lose to someone who does (watch from 2:20). After years of claims about what AI can do, will do and secretly already does, people are tired, and that fatigue is the opening for “AI agents can run your whole business now” (watch from 2:40).

The Jarvis report

The centrepiece is a viral format: a creator says hello to a voice assistant styled after Iron Man's Jarvis, and it reads back a weekly report. The report covers app downloads and revenue, short-form posting cadence, ad spend and return on ad spend, customer emails resolved automatically, and a feature request passed to a “developer agent”, whose pull request now awaits review. Because the original used copyrighted music, Morrison voices the assistant's lines instead (watch from 3:00).

The critique lands on four points (watch from 4:20; 5:00; 5:20):

• Nobody wants to talk to their computer. Morrison points to the cool reception for pitches of a computer you talk to all day. A dashboard shows the same numbers faster than a spoken report, and for a fraction of the running cost.

• Customers want humans. Do the customers know their emails were answered by an AI, and did the agent actually understand what they wanted?

• The interface is theatre. The best monitor in the room shows a pulsing blob and a radar sweep, the least useful part of the whole system.

• The report looks staged. The same creator posted another viral video in which the agent delivered the same report word for word. The content the agents supposedly plan turns out to be the same handful of video ideas, made over and over (watch from 5:40).

Escalation

The genre then escalates. A second creator's assistant reports millions of views, 64 short-form videos a day across more than a hundred Instagram accounts, and over a hundred AI-made “UGC” ad assets (watch from 6:20). Morrison's verdict is that, even if real, this is not an agentic flex. It is a display of spending, and a factory for the low-effort content that crowds out human work online (watch from 7:20). A third version runs locally on a stack of consumer GPUs with AR glasses for control. Morrison imagines its owner, years later, still unable to say what the glasses do that a phone could not (watch from 8:20; 9:00).

The finale is Morrison's own Jarvis, and its report is brutally honest (watch from 10:00):

• Fifty videos that reached under a thousand views.

• An inbox full of marketing from other agentic entrepreneurs.

• Two downloads, both from relatives.

• A financial agent advising a real job, because the credit cards are maxed out paying for the agents.

• Then it hits its token limit and says goodbye.

What the joke gets right

The closing line is the serious one: agents are cool and useful, just not in the way social media sells them. Use them for a clearly defined reason, not out of fear of being left behind (watch from 11:00).

That suggests a simple test for any agent demo, ours included:

• Measurable outcome. Does it name an outcome you can check?

• Fair baseline. Does it compare against the boring alternative, such as the dashboard, the script or the human?

• Full cost. Does it count the cost in tokens and hardware?

• Honest with people. Is it honest with the people on the other end?

That last point is not only etiquette. If the “street interview” UGC ads in these reports show AI-generated people presented as real customers, they run into the United States Federal Trade Commission's rule banning fake testimonials, including AI-generated ones, discussed in our comfyUI post on AI video (Federal Trade Commission, 2024).

References

Federal Trade Commission. (2024, August 14). Federal Trade Commission announces final rule banning fake reviews and testimonials [Press release]. https://www.ftc.gov/news-events/news/press-releases/2024/08/federal-trade-commission-announces-final-rule-banning-fake-reviews-testimonials

Morrison, E. (2026, July 10). agentic ai cringelords [Video]. YouTube. https://www.youtube.com/watch?v=d_H9VuYD8IY