AI isn’t taking your job, it’s taking control of your workday

AI is often sold as a force that will wipe out jobs and automate everything. But for many workers, that’s not what’s happening. Instead of replacing them, AI is quietly changing how they’re measured, managed, and pressured every day.

One of the clearest examples of this shift is the rise of the “token budget” – a new way for companies to track how much AI you’re using, and then use that number as a proxy for your productivity. It’s less about machines taking your job, and more about machines taking control of how you do it.

What is a token budget?

Most modern AI tools, especially large language models (LLMs), are priced and measured in “tokens.” A token is basically a chunk of text – a few characters, part of a word, or a short word. Every time you use an AI model to write, code, or summarize, you’re spending tokens.

A “token budget” is when a company sets expectations or targets around how many tokens employees should burn through. In some organizations, this has turned into internal dashboards and leaderboards showing who is using the most AI, with the assumption that more tokens = more productivity.

On paper, it sounds like a way to prove AI is being adopted. In practice, it can become a system that rewards volume over value – pushing people to generate more AI output, even if it’s low quality or creates more work later.

The rise of slop leaderboards

Some large tech companies reportedly track employees on internal “token leaderboards.” The people at the top are the ones burning the most tokens, churning out huge amounts of AI-generated text or code. That output is often called “slop” – content that looks like work, but isn’t necessarily useful.

The problem is obvious: the more time you spend prompting AI to spit out content, the less time you have to review, refine, and truly understand what you’re shipping. The person at the top of the leaderboard might be producing the most slop, but reviewing the least actual work.

For managers and executives, though, this is tempting. They finally have a simple, trackable number that looks like productivity. It’s the same old mistake as counting lines of code – except now it’s tokens instead of text.

Why the “AI will take your job” story is so powerful

The constant drumbeat that “AI is coming for your job” isn’t just a prediction. It’s also a powerful negotiating tool for companies and AI vendors.

If workers believe they’ll be obsolete in a few years, they’re more likely to accept lower pay, skip asking for raises, and cling to whatever job they have. The fear of replacement pushes people to tolerate worse conditions because they feel they have no leverage.

At the same time, big companies can tell investors they’re aggressively adopting AI, cutting costs, and becoming more efficient – even if, internally, the tools are clunky, unreliable, or barely used. The narrative sells, even when the reality is messy.

This tension shows up across the AI economy. For a deeper economic look at how AI might affect wages and employment, see this breakdown of whether AI will destroy jobs or raise wages.

AI isn’t working as well as the hype suggests

Inside many companies, AI is far from a magic bullet. Teams struggle to integrate tools into real workflows, keep data secure, and get consistent results. Even nimble startups that live and breathe technology often admit they don’t really know how to make AI reliably useful across the whole company.

If small, focused teams are struggling, it’s hard to believe that huge enterprises with tens of thousands of employees and complex legacy systems have it all figured out. Yet the public story is often the opposite: AI is everywhere, it’s working brilliantly, and any company not spending aggressively on it is falling behind.

That gap between story and reality is where token budgets and usage metrics come in. They give leaders something concrete to point at – “Look how many tokens we’re using per employee!” – even if the impact on real outcomes is unclear or negative.

The “Bittar lesson”: why AI struggles with precision

Underneath all of this is a technical limitation that often gets glossed over: large language models are great at approximations, not precision. They’re essentially extremely advanced autocomplete systems. They predict the next likely word based on patterns in data, not a deep understanding of your exact intent.

That leads to what we can call the “Bittar lesson”: the more precision you need, the less useful AI becomes.

Here’s why:

• LLMs approximate language.
• Language itself only approximates what we really mean.
• Then that approximate language is used to generate code, designs, emails, or documents.

By the time you get to the final output, you’re several layers removed from your original intent. You can guide the model and refine prompts, but you rarely get perfect alignment. AI often gets you 80% of the way there – which sounds great – but the last 20% is usually the hardest, most critical part.

That final 20% still requires human judgment, domain knowledge, and responsibility. In many cases, that’s where the real value (and risk) lives.

AI that adds work instead of removing it

When companies chase token usage and AI volume, they often create a strange outcome: AI doesn’t reduce work, it multiplies it.

Employees now have two jobs:

• Do their actual work.
• Feed, prompt, and clean up after AI systems – while being judged on how much they use them.

Instead of freeing people, AI can become a second layer of bureaucracy. You generate drafts, then spend time fixing hallucinations, checking facts, and rewriting awkward text. You get code suggestions, then spend extra time reviewing and testing because you can’t fully trust what the model produced.

This dynamic is already visible in coding. AI coding assistants can speed up boilerplate and routine tasks, but they also risk encouraging shallow understanding and fragile code if teams lean on them too heavily. For a closer look at that trade-off, see this analysis of why AI coding works – and why that’s exactly the problem.

How AI fear suppresses wages and negotiation

One of the most subtle effects of the current AI narrative is on pay. When workers are constantly told that AI can do their job, it becomes harder to ask for more money or better conditions.

Imagine asking for a raise and hearing, “You know AI can do your job, right?” Even if that’s not fully true, the threat is enough to make many people back down. Some may even accept pay cuts or heavier workloads just to feel “safe.”

In reality, for most knowledge work today, humans are still far more valuable than AI. The tools can assist, but they don’t own responsibility, context, or accountability. Yet the fear that they might someday replace you is already being used as leverage.

Who actually wins from slop budgets?

Token budgets and AI usage targets are a dream for certain players:

• AI vendors win because companies feel pressured to spend more on models, tokens, and infrastructure.
• Executives win because they can tell investors they’re “all-in on AI” and point to impressive usage stats.
• Middle management gets a new metric to monitor and control employees.

The people who don’t clearly win are the workers. They’re the ones juggling their real responsibilities with new expectations to constantly use AI, hit token targets, and prove they’re not “replaceable.”

And customers don’t necessarily win either. They don’t care how many tokens were burned to create a product or service. They care whether it works, feels thoughtful, and solves their problem.

Slow, human-centered work will outlast slop

Despite the hype, the companies that obsess over token usage and AI volume are likely to lose in the long run. Chasing slop and dashboards is easier than doing the hard work of understanding customers, building quality products, and making careful decisions – but it doesn’t create lasting value.

The real advantage will belong to organizations that:

• Use AI selectively, not compulsively.
• Keep humans in charge of the hardest, most precise decisions.
• Prioritize quality over quantity, even if it means slower output.
• Refuse to measure people purely by how much AI they trigger.

AI can be a useful tool, but it’s not a replacement for judgment, craft, or responsibility. When it’s treated as a volume machine to impress investors or squeeze workers, it becomes less about intelligence and more about control.

Why speaking honestly about AI at work matters

Right now, the loudest story about AI is still the bullish one: massive productivity, huge cost savings, inevitable job loss. What’s missing are more honest accounts from inside companies about what’s actually working, what isn’t, and how much extra friction AI sometimes adds.

Workers and teams who share their real experiences – anonymously if needed – help rebalance the narrative. They make it harder for token budgets and slop metrics to be sold as unquestioned progress. They also help others feel less alone when they’re struggling to make AI genuinely useful instead of just performative.

AI isn’t just reshaping technology. It’s reshaping power inside workplaces. The more clearly we talk about that, the harder it becomes to hide behind dashboards, leaderboards, and token counts.