AI Hype Is a Coping Mechanism

July 26 wrapped up in a somewhat scary way. On the 28–29th, an open letter named "Pacing the Frontier" signed by 1000+ employees of the top US AI companies was published. In this letter the signatories request that the U.S. government support international efforts to develop the technical and governance tools necessary to safely pace frontier AI progress.

Before this letter was published, recent blog posts both from OpenAI and Anthropic explained how GPT and Claude were being used to "improve themselves". At first, it was hard to gauge the weight of those claims, but seeing this letter confirmed those weren't just routine announcements. That made me wonder: the publications were clear and easy to understand, so why didn't I take them seriously before the letter? The answer is that my brain learned to cancel the noise of the AI hype to which it's constantly being exposed. But this time, it overcorrected and I let one of the most important things to happen in recent years pass by without a second glance. In times like these, having a clouded judgment is especially dangerous, and I'm sure I'm not the only one experiencing this. In this small piece I would like to discuss this problem that many seem to be underestimating. I am an engineer in the AI industry, and in what follows I'll try to present things from the POV of someone watching the results of their job get instrumentalized on a daily basis.

I. Engineering means something

Growing up, engineering felt like magic to me, and I always wanted to be part of it. Being able to bend physics to create technology is truly amazing. Take a look around you, no matter the object you pick I can guarantee that an engineer was involved in its making. There are so many things to get excited about that we've simply accepted as normal. Can we take a step back and talk about a few of them?

• Electromagnetism: have you ever looked at the Maxwell equations? To go from there and achieve "normalized" wireless interconnection is hard to believe.

How often I think about this : 3–4 times a week.

• Cars: Just the concept, without even going into the details of what engineering marvels modern cars are. We explode oil inside metal boxes to make wheels turn, yet a 2-ton sedan feels like a cloud to ride and maneuver.

How often I think about this : 1–2 times a week.

• Transistors: Semiconductor components that control or block the flow of electricity. Somehow we managed to make them smaller than viruses, stack billions of them on a board (CPU) in a way that makes sense, resulting in all this modern technology. The most impressive thing is the speed at which computing power has progressed (Moore's Law if you want to learn more about this topic).

How often I think about this : ~5 times a week.

• Honorable mentions: planes, nuclear reactors, disk storage, cameras, audio devices, internet and many more. Heck, they even sent humans to the moon with 4KB of RAM.

To help you get an idea of the scale of this accomplishment: today any budget computer has at least 8GB of RAM, i.e. 2 million times more.

The latest marvel, and our main focus here, is AI. To make things clear, what I mean by AI is the mainstream meaning given to it: things related to LLMs and generative models. I could argue that even vanilla machine learning is impressive (Boosting* is still the production king for forecasting), but that's not sexy enough for our current timeline.

*Boosting is a family of decision-tree machine learning algorithms. It represents a classical statistical approach that predates the modern generative AI and LLM boom.

I distinctly remember finding Talking Ben the dog very impressive the first time I used it, so I can't begin to imagine how kids felt when ChatGPT dropped in 2022. I can't stress enough how incredible it is that, using linear algebra and clever engineering, we nearly solved language from a technical standpoint.

I do think AI is one of the biggest revolutions the world has ever known, and that's precisely why it's really important to keep questioning ourselves as things move forward. When agents* were introduced, it felt like a natural evolution, and ever since models like Opus 4.5 and GPT 5.2 arrived, things clicked into place. Those models were finally smart enough to use tools to complete actual work thanks to their ability to pursue long horizon tasks. But, along the way, things had started going wrong.

*An agent is a LLM model that has access to tools to complete tasks assigned to it.

II. How we're being misled

The replacement narrative and tool-hype aren't new. The 9–9–6 Silicon Valley doomers* have been harassing us for years now. You haven't mastered this tool? You're getting replaced. You still do this by hand? Empty your desk. You're not tokenmaxing**? Get ready for permanent underclass. If you've been scrolling on LinkedIn (or even worse, Twitter), any new thing coming out is a revolution. Everything is a game changer, and we're hitting the sainted AGI*** moment every week. So here are 3 genuinely cool things but that consistently got oversold and presented as next-gen engineering:

A. Skills

Skills are procedural instructions that a model can decide to load or not depending on the task it was asked to complete. The clever idea is the mechanism of progressive disclosure that loads information in layers into the context to avoid overcrowding it. But fundamentally what is a skill? A text file or a collection of them. Literally just an instruction file or code snippet that the model loads into context. So while those are great for improving model performance on tasks without having to respecify instructions every time, I wouldn't call this a revolution.

A hard truth about skills, and prompt engineering in general, is that it represents a massive technical debt. Models' behaviors vary wildly, and a skill tuned for Opus 4.5 might perform poorly on Opus 4.8. Having to rewrite all your skills at each model drop represents a massive bottleneck.

B. Agent loops

Agent loops are a new way of working on long horizon goals. You set a goal and success criteria and then let the agent orchestrate everything, often for hours. The agent decides which tasks to execute, auto-evaluates its progress against the success criteria, and stops only when the end goal is met. So what is an agent loop? A glorified while loop****. Instead of evaluating a simple boolean condition, the loop runs until a user-defined goal is satisfied. I know this requires solid software orchestration, but come on, we split atoms to generate electricity.

C. Claude Tag

This is the most recent one. It's a new way of interacting with Claude. Essentially Claude becomes a coworker in slack that has company knowledge. It knows what items are in the backlog, can work on said items, can autonomously message you about them and can interact "normally" in any group chat. This is great, and I'm sure it will speed up a lot of daily tasks. But, once again, what is this innovation at its core? It's a (very clever) way to manage the context available to Claude to help it understand what is happening. It leverages a few clever MCPs***** to make interactions feel seamless, but at its core, this is still just context management.

*The '9–9–6 Silicon Valley doomers' are people who believe the future belongs to those who work relentlessly (9 a.m. to 9 p.m., six days a week) and who often hold pessimistic or extreme views about AI, competition, and the need to sacrifice work-life balance to avoid being left behind.

** Tokenmaxing refers to aggressively consuming, generating, or optimizing LLM usage to maximize productivity or show off AI adoption.

*** AGI (Artificial General Intelligence) is a hypothetical stage of AI development where a machine possesses human-level (or superior) any intellectual task.

****A while loop is a type of structure that executes a block of code as long as a specified condition remains true

*****MCP, or Model Context Protocol, is a standard open protocol that enables the models to use external tools.

III. Conclusion

Maybe I am being hypersensitive about this matter, and it probably shouldn't annoy me as much as it currently does, but I truly believe this constant hype about things that don't deserve it (that much) contributes greatly to the overall FOMO-y anxious vibe. I actually think that the people that keep this train going are the ones who are the most afraid. In reality, acting like they do (hyping anything that they hear about) is a coping mechanism. By convincing themselves that they're at the forefront of the next revolution, they believe they won't be replaced by AI, unlike us peasants who don't use the latest prompting framework. The AI labs managed to frighten them so much that they transformed into the best salespeople ever, running free advertisements on a daily basis for them. Looking back, many of the pseudo-revolutions that happened in the AI landscape aren't relevant anymore today. Remember when you had to do RP with a model to get something out of it? Even better: remember when you had to threaten to kill yourself to have GPT answer your requests correctly? That was a revolution.

I think people keep falling for the rhetoric of the labs, forgetting that they're businesses. The labs use marketing too, you should use your brain to defend yourself against it. Even the companies that adopted AI the hardest cannot directly link their AI usage to any significant gains. Uber, which burned all its 2026 AI budget in 4 months, seems to be questioning whether AI is worth it at this scale. Labs claim AI makes workers 10x more productive, yet the only entities that have 10xed their valuations are the labs themselves.

Genuinely cool things are happening and are being built, and we're so lucky to have access to all these innovations. However, those engineering achievements are being hijacked by engagement-farmers that spread misinformation and scare people. AI is scary enough on its own and requires vigilance as we transition the world toward it. The "Pacing the Frontier" letter is the proof that today, more than ever, AI is a subject that should concern us all. The self-appointed commentators that took over the public discourse do far more damage than they realize. By exhausting the world through their bullshit claims, they are effectively precluding people from accessing real information. AI is my job, yet I find it harder every day to navigate the ocean of slop we're served.

If you have someone in your circle that can help you understand those innovations, ask them as many questions as you can, it's important. If you don't have that someone, hit me up! Just don't let the clout-chasers scare you, make you any more anxious than you are or make you lose interest in the subject. AI is here to stay, and it's all our responsibility to make sure it makes our lives better, not worse.