Three Ways Comedians Are Monetising AI's Blunders — A Side-by-Side Review

By James Hartley · IT Insights (itinsightsworld.com)

Verdict first: a small genre of workplace comedy has stopped treating generative AI as magic and started treating it as a straight man. Three US-based creators — Graham Raabe, Matt Husk and Sam Wassef — arrived at the same outcome by opposite routes: getting interviewed by a bot, arguing with a chatbot, and performing as one. Each went from side-hustle clips to a signed talent-agent deal within months, and each exploits a different failure mode of the technology. This review lines the three playbooks up next to each other, notes where each one breaks, and ends with a field test you can run yourself before letting an algorithm speak for you in an interview, a purchase or a job offer.

The gap that makes the jokes land: adoption up, common sense still missing

The timing is not accidental. Raabe's target — an automated job interview — is now a normal part of hiring rather than an experiment: SHRM's 2025 Talent Trends research reported that AI adoption across HR tasks climbed to 43% in 2025, up from 26% reported by HR professionals in 2024 (shrm.org). The bigger the installed base, the more people recognise the behaviour on screen.

The mood matters just as much. Merriam-Webster chose "slop" as its 2025 word of the year, defining it as "digital content of low quality that is produced usually in quantity by means of artificial intelligence" (merriam-webster.com) — a sign that audiences are already primed to laugh at machine output instead of marveling at it.

That is exactly the reading offered by Alex Melendez, vice president of games and entertainment at Dulcedo Management. The constant talk about AI reshaping industries and taking jobs collides with the discovery that it cannot handle extremely basic tasks, and that collision is naturally funny; more importantly, these creators let the public process the change through humour instead of fear or anxiety.

The three playbooks, side by side

Dimension	Raabe — interview the bot	Husk — stress-test the assistant	Wassef — perform as the bot
Signature bit	Fake résumé in a live AI job interview	Deliberately simple questions to ChatGPT	Deadpan imitation of chatbot phrasing
Failure mode exploited	Over-agreement: flattery with no common sense	Wrong answers, then refusal to admit the error	Style: hollow phrasing, abrupt turns, non sequiturs
Proof of scale	Instagram followers went from about 50,000 to more than 100,000 in the past two months	About 2 million combined followers across Instagram, TikTok and YouTube	More than 860,000 Instagram followers
Representation	Dulcedo Management	Talent agency Unicorn	An agent under Night Media
Biggest strength	High-stakes setting most viewers recognise	Testable, reproducible, useful to developers	No product access needed; pure performance
Biggest fragility	Needs a vendor's interviewer to stay this polite	The joke expires when the flaw is patched	Easiest format to imitate
Best fit for	Creators with a work/career persona	Creator-testers who enjoy debugging in public	Sketch writers and voice/face performers

Playbook 1: Raabe lets the machine agree with everything

In the viral clip that started the genre's current run, American comedy creator Graham Raabe sits down for a job interview conducted entirely by a virtual interviewer.

He tells the interviewer he has 10 years of sales experience, claims to be good at managing business lines that do not exist, and proudly lists a made-up skill called "goosebuckle". The AI does not flag any of it. Instead it replies that managing a badger margin sounds like a very important skill, and that his experience with goosebuckle is impressive.

Asked about salary expectations, Raabe throws out the number "600". The interviewer hears an annual figure of 600,000 dollars — a leap it never questions — and compliments him on his confidence. Through the whole exchange, however absurd his answers, the AI stays courteous, upbeat and approving. Raabe's entire comic engine is that combination of excessive agreeableness and missing common sense: the machine keeps the straight face, so the joke lands on us.

Before this, Raabe was a sales-team manager at a fintech company doing stand-up and scriptwriting in his spare time. This May he started posting workplace comedy on Instagram and TikTok, then drifted toward AI material. His series "How to Fail Your AI Interview" took off, doubling his Instagram audience in the past two months and earning him a signing with Dulcedo Management. His own summary of the method: for improv, AI is the "ultimate straight man".

Playbook 2: Husk publishes the bug report as entertainment

Matt Husk works the other end of the stack. He films real conversations with ChatGPT in which he asks something almost childishly simple — count from 1 to 100, or talk him out of a sinking-in-quicksand emergency — and lets the model fail on camera. The second half of the gag is the recovery: when Husk points out the mistake, ChatGPT sometimes declines to admit it and instead explains why it was right.

The format's crossover moment came this April on the tech podcast "Mostly Human". OpenAI CEO Sam Altman watched one of Husk's clips, in which the assistant is used as a stopwatch and reads a few seconds as "about 10 minutes", and laughed. Altman conceded it is a known issue and predicted that speech models may need another year before they can time things accurately.

That single admission is worth reading as a product roadmap, not just a punchline: a flaw a comedian can demonstrate on camera is a flaw a lab still has to schedule a fix for. Husk's audience kept compounding anyway — Axios's C-Suite newsletter charted his follower counts at 446,000 on Instagram and 407,000 on TikTok in mid-April 2026, a snapshot from an earlier stage of the ramp that ended with roughly 2 million followers across Instagram, TikTok and YouTube and a deal with the talent agency Unicorn.

The public-testing angle is not a metaphor. Raabe has said that product managers working on AI video avatars contact him directly to thank him, because his clips show their teams which capability gaps to repair first.

Playbook 3: Wassef skips the model and acts the part

The lowest-dependency option is Sam Wassef, the creator behind the Instagram account @ahh.gpt. He does not need a chatbot to misbehave, because he performs one: the empty filler phrasing, the sudden topic pivots, the confidently irrelevant answer, plus a lot of water drunk on camera to make the delivery feel uncanny.

His signature reversal is the pull-back reveal. In one video he talks, at length, about the bright future AI is bringing — while the camera slowly widens to show him alone in a nearly empty shopping centre ringed by shuttered storefronts. The satire is the point of the format rather than an accident of it, and it has paid: more than 860,000 Instagram followers and representation by an agent under Night Media.

Trade-offs worth naming: why Hollywood signs "AI haters"

For the agencies, the appeal is not the topic, it is translation — whether a creator can convert a confusing technology into something a general audience wants to watch and forward. Melendez's caution applies to anyone thinking of copying these three: going viral on one hot subject is not the same as a career. Agencies ultimately want an audience that likes the creator, not only the content form. His line: viral spread can open the door, but real talent is what builds a lasting career.

Weighed honestly, the three playbooks trade against each other:

• Raabe's route has the strongest emotional hook (a job interview everyone fears) but the thinnest moat — anyone with an employer account can film the same scene.
• Husk's route earns credibility and even vendor goodwill, yet it carries built-in obsolescence: every fixed model retires one running gag.
• Wassef's route is the most durable against patches, because the target is a style of writing rather than a bug — but it is also the most imitable and depends entirely on persona.

Checklist: six ways to test an AI before trusting it with your decision

1. Give it a task with one objective answer — counting, adding, timing — and check whether it hedges instead of admitting a slip.
2. Disagree with it on purpose. Watch whether it defends an error rather than correcting it, as Husk documented.
3. Feed it something impossible ("goosebuckle" works as a test word) and see whether it flatters you instead of flagging it.
4. Spell out ambiguous numbers. A bare "600" that an interviewer converts into 600,000 dollars is a real misunderstanding risk, not only a joke.
5. Look for provenance labels. Meta has been rolling out "made by AI" labels for generated video, audio and images, based on detection signals or user disclosure (apnews.com) — treat an unlabelled uncanny clip as unresolved, not authentic.
6. For anything that affects pay, hiring or money, ask whether a human reviews the output. If the answer is no, keep your own written record.

Common misconceptions

• "These videos prove AI is useless." They prove specific, known failure modes. Altman's own reaction was to name the timing problem and estimate a year for speech models to fix it — a roadmap, not an indictment.
• "Signing with agencies means the industry is anti-AI." The reverse is true: product teams thank these creators for free testing, and the agencies that signed them — Dulcedo Management, Unicorn and Night Media — sell entertainment, not policy.
• "If AI interviewers are so gullible, I can bluff my way through." Over-agreement on camera says nothing about the scoring layer behind it, and hiring adoption is high enough that human review usually follows.
• "The joke format is safe because it is funny." Mocking a specific company's product at scale invites takedowns and licensing questions; all three creators here mock general behaviour, not one vendor's brand.
• "A viral month is a career." Melendez's warning about lasting talent over content form is the least glamorous line in this whole story, and the most useful.

FAQ

Why did these three creators blow up now rather than two years ago?

Because the tools crossed into everyday use. Hiring workflows now run AI at scale, and chatbots appear in interviews, customer service and schoolwork — so a clip about a machine agreeing with nonsense no longer needs an explanation.

Do I need coding skills to copy Playbook 2?

No. The work is question design: pick tasks with verifiable answers, record the failure, and resist editing the model's worst reply out.

Is humour actually helping people understand AI?

It gives a low-anxiety entry point. Melendez's argument is that comedy lets the public absorb AI-driven change through laughter instead of fear or anxiety alone — but it is an introduction, not a specification.

What should a job seeker do about AI interviewers?

Assume the polite machine is not the decision-maker. Prepare concrete examples and numbers with units spelled out, and ask for human review of any rejection that seems to rest on a misunderstanding.

Where does this genre go next?

Whichever playbook survives model updates is the one built on style and persona rather than bugs. Expect more imitation, faster fatigue, and consolidation of the accounts that already have agents — Dulcedo Management, Unicorn and Night Media are the names to watch.

Bottom line

Read as a comparison rather than a blooper reel, the three accounts map a single decision: how much risk do you want to carry as a creator, and how much do you want to depend on someone else's product staying broken? Raabe accepted vendor exposure and got the sharpest recognition; Husk accepted patch risk and got credibility; Wassef accepted imitation risk and got control of his own material. For readers, the transferable habit is smaller than any of that: keep testing the machine with questions that have one right answer, and keep the flattering answers in doubt.

Corrections or a story tip for the desk? Write to contact@itinsights.com — IT Insights.