Insights for Growth

Stuck in the Middle With You: Work, Our Agentic Overlords, and How the Digital Creative Sausage Is Made

It’s the fever dream of some billionaires, like Jack Dorsey, chairman of Block, Inc., to position AI to replace human middle managers in many, or most, organizations. They imagine that the company of the very near future will be structured and run as a ‘mini Artificial General Intelligence (AGI),’ a super-smart, super-competent replacement for the functions of middle managers. The ultimate goal is to collapse the organizational structure into a singularity of efficiency and revenue growth.

Unfortunately, there will be no interview with “The Bobs” in the slim hope of keeping the job you already have. That job is going, going…gone. It’ll be a cataclysmic flattening of the management structure where much of what managers used to do gets in-sourced to AI agents that channel information to where it needs to go, generate reports, and make most decisions autonomously. Humans will still be there (so the billioned tech hegemons say), but there will be a lot fewer of them, and they’ll mostly focus on the people-y things AI doesn’t do well: coach, mentor, create consensus, foster relationships. Which I guess is good unless you’re the sort of manager who likes doing process-oriented things. And heck, we also know a lot of managers who aren’t even good at the people-y things, right?

It’s also been well-reported how the various flavors of artificial intelligence are well on their way to erasing job opportunities for junior and entry-level professionals. For example:

• A report from the World Economic Forum says U.S. entry-level job postings fell 35% in the 18 months before March 2026.
• Forbes reported a 13% employment drop among ages 22–25 in some AI-sensitive jobs.

With the relentless drive from executives at the top to flatten the organization into AGI agentic efficiency and general-purpose AI at the bottom ready to send out “Don’t Apply Here” messages to recent college grads, it feels like someone’s fired up a Spotify playlist of “Stuck in the Middle with You” on repeat while Mr. Blonde gleefully douses middle management with gasoline. (I’m a sucker for Stealers Wheel, but, man, Reservoir Dogs left me scarred).

But even if you do manage to land a job at FutureCorp, there’s a surveillance workplace waiting for you. It’s designed to force you, Mr. Employee, to unwittingly train AI to replace you with a more efficient version of you that doesn’t need sleep, bathroom breaks, or health benefits (though it will be needing all of your electricity, thank you). Take Meta’s controversial Model Capability Initiative (MCI). It records employee mouse movements, clicks, keystrokes, and screenshots, all to train AI models about how people work. You can bet Meta won’t be the last to do this (they might not even be the first). There is a certain perversity to participating in the destruction of one’s own livelihood, but it is understandable: we all need to eat today. The existential threat of unemployment is for tomorrow.

As much as I’m intrigued by the technology and excited by the possibilities it unlocks, I am concerned about the ways AI is hollowing out the creative profession. So it was with curiosity and dread that I ended up as a part-time AI trainer, helping LLMs to understand the entire chain of knowledge work.

Making the AI Sausage

I needed to learn how the digital sausage is made, so to speak. And boy did I. It came in two main flavors: improving generative models and simulating a creative workflow.

1. The basics: Improving generative models. The simplest work involved annotating images so that generative AI models could understand what they were “seeing” in an image and be able to reproduce it when prompted to do so by a user. All items in an image had to be tagged with identifying labels to build a library of examples the AI model could draw upon. The work also involved grading Generative AI image outputs against a pre-written prompt. Did the model follow all of the instructions in the prompt? Were there weird “uncanny valley” effects that gave the image away as AI-generated? Was it visually accurate? Every rating choice needed an explanation.
2. Simulating an agentic creative workflow. For some tasks, the purpose was to act in the role of senior manager and generate a work strategy prompt, as if instructing an employee on what to create. In other tasks, the goal was to be a junior manager and develop a rubric, a scoring guide, to objectively measure how well the employee completed the ask. And finally, there was the employee output: designing an end product based on the directive (prompt) supplied by the manager using open-source creative tools. Ultimately, the work of hundreds of people on these training tasks will be used as learning materials to improve AI agents that may, one day soon, be able to reproduce a full creative workflow without humans.

The experience was both fascinating and uncomfortable. I have conflicting feelings about participating. But it seems that I’m in good company: articles in Wired and elsewhere tell of everyone from marketers and programmers to Hollywood screenwriters who are now side-gigging as AI trainers teaching LLMs how to take their jobs.

Sometimes I feel a bit like a traitor to my profession doing these projects; on the other hand, I see it as competitive research and a hedge against obsolescence: How can I learn what I’ll be up against as models get better at marketing and creative workflows? Where can I plug myself into technology frameworks where I can add value, and thus, hopefully, maintain relevance, boost my creativity, and uncover niches the machines can’t fill?

AI companies want to understand the entire knowledge work value chain so they can automate it away from the masses in middle management. And as AI eats up entry-level jobs, we have no idea how younger generations of workers will gain career experience. It looks like we’re all stuck in the middle for now, and hopefully someone will find a solution before Mr. Blonde throws a match on all that gasoline. My optimistic sense is that we may yet get beyond all the doom and gloom and discover that the irreplaceable role of humans is our unique ability to craft stories driven by emotion, enabled and enhanced by AI tools, not replaced by them.