AI can't replace jobs

It can automate some tasks - this distinction is very important

The title/subtitle really says it all here. There is one exception: if a job is little more than a single task, then sure - AI can probably replace this job. If a job was that simple, should it have ever been a job in the first place?

Note: Most of this was written in July 2025 and I thought it was too late to put it out, but Citrini Research’s fantasy fiction piece, The 2028 Global Intelligence Crisis convinced me it could still be useful. There seems to be a fundamental misunderstanding here around how work works. Jobs, tasks, and work are all very different things - related, but different.

Simple Jobs

Simple jobs have always been around and they’re always getting replaced. Human beings were once employed to walk around and light streetlamps when it got dark. Electricity and light sensors replaced them. At one point, every manager and executive had secretaries to type for them. Put a PC on every desk and now only execs at the highest level can still justify an executive assistant or chief of staff.

When I started out in IT I worked at one of the world’s largest payment processors. I had many roles before I got into cybersecurity proper, but automation was one of my favorite.

Automated jobs were scattered throughout the organization. They were written in Perl, Visual Basic, Bash, C++, Crystal Reports. The individual that originally automated the task chose the language they knew best. Some of the jobs ran on servers, but most existed on someone’s personal computer, or a secondary computer under their desk, in their cubicle.

A few employees convoluted these tasks so much, that they became full-time jobs. Many weren’t professional developers, so the concept of monitoring, alerting, logging, and error handling didn’t occur to them. Their code was just bad enough that they needed to babysit it every single day and step in when things broke.

We bought a commercial automation platform called, AppWorx, and it was my job to centralize and normalize all these tasks in one place. It was hugely fulfilling work - I’m the flavor of neurospicy that loves to dive into a mess and organize it. The part that sucked was putting several of my colleagues out of work.

I owned this Think Geek shirt back then. It was hilarious until it became a little too real.

Bullshit Jobs

I read the book Bullshit Jobs by David Graeber last year and it was a timely revelation. I found it interesting that the definition Graeber goes by is self-defined. When workers believe that their own jobs shouldn’t exist, it is a bullshit job.

Graeber generally states that he found about half of societal work to be pointless - sometimes this is part of a single job (e.g. someone might feel that 70% of their job is pointless, but the other 30% worthwhile), or that the role entirely shouldn’t exist. He came up with five categories for bullshit jobs (copied straight from the Wikipedia page linked above).

1. Flunkies, who serve to make their superiors feel important, e.g., receptionists, administrative assistants, door attendants, store greeters;

2. Goons, who act to harm or deceive others on behalf of their employer, or to prevent other goons from doing so, e.g., lobbyists, corporate lawyers, telemarketers, public relations specialists;

3. Duct tapers, who temporarily fix problems that could be fixed permanently, e.g., programmers repairing shoddy code, airline desk staff who calm passengers with lost luggage;

4. Box tickers, who create the appearance that something useful is being done when it is not, e.g., survey administrators, in-house magazine journalists, corporate compliance officers, academic administration;

5. Taskmasters, who create extra work for those who do not need it, e.g., middle management, leadership professionals.

The folks I automated out of a job fell firmly into the Duct Tapers category. The tasks they partially operated worked, but not well enough for them to move on to another task.

I clearly recall coming to this job one day, walking through the doors, gazing across the cubicles, and having a revelation: two-thirds of the people that work here could never come to work again and there would be zero impact. In each department within the company, I had observed that there were one or two ‘heroes’ that seemed to keep everything working smoothly. The remaining members of the team either managed something small and basic, or managed something that didn’t really need to be managed at all. They engaged in a sort of work theatre, as if Fisher Price made enterprise-grade playsets for storage administrators and backup management.

The other big revelation from this book was that not all jobs exist because work needs to get done. There are vanity hires - the flunkies mentioned in the first category above.

Companies are naturally inefficient

New or smaller bootstrapped companies are loathe to spend or hire too much, unless it’s really necessary. As companies get larger, it becomes more and more difficult to understand what is necessary or not.

Managers say they need more people, so they get them. It’s literally part of their job - to manage people. Asking a manager to determine if they need more people is almost a conflict of interest - of course most will say yes. Even if they believe that they legitimately need more staff, can the productivity improvements be measured? Can the hires be justified? They often can’t, which is why large rounds of layoffs every few years is necessary to compensate for this inability to measure productivity.

If you handed your average information worker pen and paper and asked them to categorize how they spend a 40 hour work week, they might be challenged to do so. What category is Slack and Email? Is it wasted time, or is it productive? Sixteen hours in meetings - were they all necessary? Did they all need to be an hour long? Could you have skipped half of them and done something more productive instead?

If the individual struggles to answer this question, you can bet the company doesn’t know any better. That’s why sometimes layoffs have zero impact on the company and others result in a clear drop in performance or quality that customers notice - it’s often guesswork.

Is AI replacing jobs?

It’s easy for a giant tech company to lay off 10% of its workforce and attribute it to AI. AI is just the latest excuse in a long history of excuses used to give a positive spin on mass layoffs. Large companies have always done big layoffs, in part, to counter for their inability to measure employee productivity.

Management values growth and self-importance over efficiency and productivity. They’re constantly pushing for larger budgets and teams, regardless of whether it’s justified. Eventually, the company brings in a consulting firm that tells them the ugly truth: two-thirds of the company do little to nothing of value.

No one is being replaced with AI. In many of the cases where they’re very explicitly trying to replace humans with AI, things haven’t gone as well as hoped (e.g. Klarna, Salesforce).

Fallacy #1: because AI can do a task, it can replace a worker

AI can write code, therefore it can replace developers. Is writing code everything a developer does? Absolutely not. In fact, if you want AI to write code, you need more software engineers to help manage the output.

AI can create Gantt charts, so can it replace project managers?

What about cases where AI succeeds in completing a task only some of the time? There are many stories of AI inconsistency, failing 10-40% of the time - even doing the same exact task the model completed successfully in the past. That’s hardly a case where you can replace a human with AI.

Fallacy #2: AI work replaces human work

Actually, we’re finding that the reverse is often true. AI will do work that no worker was ever going to get paid to do, perhaps because it was too boring or the economics didn’t make sense. For example, I was never planning to hire human artists to create custom images for the slides I use for my talks, but now that I can have AI do it for a $20/mo subscription, it makes sense to do it.

There’s a ton of new AI—generated slop on YouTube targeting kids that certainly wouldn’t exist without AI and didn’t steal jobs from existing artists.

Again, in the case of software engineers, we need more than ever, because AI needs a human that understands the bigger picture and what the intended output should be. A human breaks down the project into tasks, writes the prompts, adjusts the prompts, reprompts when AI gets it wrong, etc.

My friend Ayman Elsawah has some great examples around the security shortcomings in AI-generated code.

The AI + Security Issue

Fallacy #3: AI hasn’t already replaced jobs

GenAI has certainly already replaced jobs. Again, these were largely tasks-as-jobs

• Contractors creating okay-ish content for marketing teams

• Contractors creating okay-ish graphic design for marketing teams

• Voice actors doing work on projects with tight budgets/margins

Conclusion

AI is certainly helping a lot of folks be more productive and more efficient, particularly in the area of software development. Will they make the workforce more efficient overall? Of that, I’m not so sure, as AI emboldens people to step way outside their wheelhouses, and it’s easy for AI to make you feel like you’re doing amazing things, while the subject matter experts looking over your shoulder are suffering retinal detachments from rolling their eyes so hard.

Take software engineering, for example. Thanks to AI, anyone can generate code. With no background in software, what will the average middle manager create? Are people with bullshit jobs now going to create bullshit software?

These folks aren’t software engineers - they’ll make every imaginable mistake when building software with AI, burning GPU cycles all the way. AI isn’t trained to intercede during a vibe coding session and say, “you’re kinda reinventing the wheel here”, it will happily burn those tokens, creating cartloads of software no one needs or asked for.

So we’ll likely end up with more jobs, more software, more tech debt, more vulnerabilities, more attack surface, more of everything, now that it can all be generated at a whim on a prompt (or launched into full auto thanks to OpenClaw and other agents).

Anyone interested in a Vibe Code Cleanup Engineer? Companies will be hiring soon.