There's a dark joke circulating in tech circles these days. Friend to friend, developer to developer, the conversation always seems to land in the same uncomfortable place: "Don't worry about your job being automated—you're an AI engineer. You create the things that replace everyone else." The punchline, according to software engineer and blogger dmanco, is that this reassurance is completely backwards. AI engineers may find themselves replaced by their own creations faster than most other developers.

What Even Is an 'AI Engineer' Anymore?

The core problem, as dmanco sees it, is semantic chaos. The term "artificial intelligence" has become so diluted that it encompasses everything from LLMs like ChatGPT to the convolutional neural networks processing your phone's camera shots, from social media recommender systems to NPC pathfinding algorithms in video games—which are often just classical A* search. These technologies share almost nothing under the hood, yet they're all dumped into the same marketing bucket. Arvind Narayanan and Sayash Kapoor tackled this exact phenomenon in their book 'AI Snake Oil,' noting that AI has become simultaneously everything and nothing. An engineer with an 'AI engineer' title could be doing anything from fine-tuning transformers to writing if-statements for game characters—and job postings reflect this confusion, offering roles so scattered across domains that searching for one means wading through irrelevance.

The Generalization Tsunami Is Coming

Here's where it gets uncomfortable for specialists. Meta's recent release of DINO—a versatile, efficient vision model adaptable to different tasks with minimal fine-tuning and no annotations—illustrates the direction the industry is heading. Foundation models are increasingly cannibalizing specialized AI branches, absorbing what once required dedicated research into general-purpose systems that work out of the box. Why hire a computer vision engineer to build a custom solution when you can plug in the latest generalization breakthrough from big tech for a fraction of the cost? Dmanco's prediction is stark: 'We'll eventually reach a point where having AI engineers and researchers will no longer be convenient for most companies.' The best talent will consolidate at major players, while the broader market becomes saturated with commoditized tooling that anyone can deploy.

Why Software Devs Might Survive Longer (For Now)

There's an irony in dmanco's analysis: software developers who integrate AI into applications may actually have more job security than the engineers who build those models. The reasoning is straightforward—someone still needs to direct the agents, verify outputs, and understand what they're constructing. 'Agents need a user, and that user has to know what they're building,' dmanco notes. This isn't something current systems can fully replace; you still need domain expertise to ensure AI does the right thing in the best way possible. But this buffer zone may be temporary rather than permanent—and it raises uncomfortable questions about how many "AI engineers" are really doing novel work versus assembling existing APIs into products.

Key Takeaways

  • 'AI engineer' has become meaningless as a job title, covering everything from LLM research to basic automation
  • Foundation models like Meta's DINO are rapidly absorbing specialized AI capabilities that once required dedicated engineers
  • The economic logic favors general models over tailored solutions for most companies within years, not decades
  • Software developers who integrate AI may outlast the engineers who build those systems—at least temporarily

The Bottom Line

Let's be real: if you've spent years becoming a "prompt engineer" or "AI integration specialist," congratulations—you've built expertise in exactly the kind of work that gets automated first. The irony isn't lost on anyone actually paying attention. Specialization used to mean job security; now it means you're a narrower target for a model update.