The all AI and jobs edition! Different than you think

PLUS 5 Trite Now and Then Comparisons, a brand new product to guess and a Menu Mistake

Apr 29, 2025

When I started this newsletter in the fall my primary intent was to have a little fun with technology stories and leaders. But as I started to investigate more I uncovered serious issues that shouldn’t be seen as humorous. So my Monday and Thursday editions will focus completely on my “take the mickey”. I will start a biweekly edition called Sunday Dinners that will take a serious look at the more discomfiting technology stories

EXPERIMENTAL RECIPES

MAIN COURSE - The all Agentic firm has become a reality. Are all our jobs doomed?

Somehow this feels like a companion piece to the robots running a half marathon I looked at last Thursday. This particular study played right to all the strengths that we have been lead to believe that Generative AI has. Navigating efficiently and effectively through the digital world, not moving physical simulacrums through the real world.

Carnegie-Mellon university conducted a research project by staffing a fake software company with nothing but digital agents. They included most well known AI models fulfilling roles such as financial analysts, software engineers, and project managers. Tasks for the various agents were based on activities typical of real software companies. This was a very elaborate project with considerable complexity in setting up a copycat company.

Too bad this company wasn’t real as I’ll be bet many of you would want to be investing in such a future corporate behemoth. I’m sure the results had to be “blow your socks off great” given the constant and never ending AI hype from purveyors of AI products and advisor hangers-on.

Dismal. Not that you can’t invest, but the overall performance of the various agents. In summary they were about as good as the running robots. Didn’t complete very many tasks. Took too long. Very costly in doing the few tasks they could. These best functioning agent - OpenHands + Claude 3.5 Sonnet - resolved 24% of their tasks. Which was twice as good as their nearest co-worker, another hapless digital agent. Feel free to use this statistic next time you have a negative performance review.

“Researchers wrote that agents are plagued with a lack of common sense, weak social skills, and a poor understanding of how to navigate the internet.” The last one really stands out. I thought the whole purpose of digital agents was to run quickly though the internet booking your lunches and buying you pet food. For those of you using digital agents I would check my Amazon purchases closely. If they have access to your credit card, an audit of transactions might be in order.

Your job is NOT lost, but these agents might be able to misplace it.

QUICKBYTES: AI candidates showing up in job competitions

In a related story it was discovered that actual humans were competing for jobs with AI generated candidates. From the previous story we know that they won’t be good enough to do the work but maybe they are great at acing the recruiting process.

Turns out that AI on its own isn’t good enough to generate fake candidates either. But you know who are good at using AI to meet their nefarious goals. Cybercriminals. They were using AI generated resumes and images as well as AI masking (seen below) in video interviews. They actually got some jobs over real candidates as their aim wasn’t gainful employment but to “steal trade secrets and sabotage a company's systems with malware.” Evidently more than one company has been hit by this scam prompting some to go back to in office interviews.

QUICKBYTES: AI can help you beat others in video job interviews

Completing a trifecta of stories about AI and jobs, is a new AI app that allows you to cheat on job interviews. (I wonder whether the cybercriminals above had that in their bag of tricks). This tool was developed by a Columbia University student (one of the few not protesting) that got him suspended as well as a cool $5.3 million from venture capitalists to build it out. Called Cluely it analyzes what is happening on the screen and audio, and makes quick suggestions on how to answer the questions posed by the interviewer.

Evidently the product has 70,000 subscribers, not sure how many are paying the $20 monthly subscription fee. In a hands-on review though the product had a number of problems typical of AI: getting the question mixed up or generating “hallucinations”. That could lead to some very funny interview interactions if you just mindlessly said what the product fed to you.

But don’t bet against this not being a big hit as this student has all the chops necessary for success: no conscience and an ability to bullshit with the best. In his words Cluelessly is a “cheating tool for literally everything”.

Once again Silicon Valley finds a huge human need to fill.