On AI FOMO, Cognitive Overload, and what actually matters in the upcoming AI era

There was quite the buzz with Matt Shumer's "Something BIG is happening" article. I read it too, just like I read Dario Amodei's "The adolescence of technology," and this provoked an in-depth internal conversation. I am allowing myself to publish this article among the other technical articles, rather than in the shards section, since it focuses more on the technical side of the problem than the philosophical one.

See, I already mentioned in a previous article that I am fascinated by AI and the opportunity emerging very quickly. And that I am a bit frightened by the quality of the end product - and this was for Opus 4.5, and now we have Opus 4.6, which is much better.

I see, however, a certain over-exageration - big names in AI, especially Dario Amodei, Matt Shumer, Nate Jones, and others are spreading some kind of cautious warning more and more often - AI comes here to take your job. Not in that direct context, but generally, that is the narrative—industries colonized by the new models, people removed from the process chain, and so on. Matt also played the "like a Covid but worse" card, which, honestly, is quite an overreaction. There is a shift in the industry. The corporations are always hungry to optimize (e.g., work with fewer people, faster) and industrialize processes.

What concerns me is the mantra: "If you don't learn AI in depth, you're done." And this applies to everyone, regardless. Everyone is pushing to spread that AI FOMO, and I feel it just like the 2021 crypto wave. I've been there and done that. I felt the FOMO, and I even burned some small money sucked into this cycle.

So, let's break down what I see as viable, and I don't actually agree with. But before that, I want to share some personal experience on the advantages, requirements, and disadvantages.

The advantages of being a power AI user

I am a power LLM user since day one - I was using AI heavily and was trying to up my game, solving our ITWS bottlenecks - primarily in the area of developing our own products during a constant flow of customer ones, also improving quality and speed of delivery for our clients, automating redundant "by-hand processes" and so on. Jumping through OpenAI chats, IDEs (Cursor, Windsurf), Cline up to the CLI era and looking for what's next.

My advantage: I can sit in the afternoon after a productive day and plan the feature requests for the projects I have in mind, then execute them at any time, and when they are done, check them.

Despite all the limitations (we'll talk about these) I am usually done with a clear set of 30-40 tasks that I can run either through my own framework developed on top of Claude (will write about in another article about it), or via Get Shit Done, or any other agentic framework (I test a lot of these on a certain group of tasks to define effectiveness level). Separately, I have several automations that do heavy lifting in marketing, advertising, and content curation.

This simply enables me to be super productive, personally using these tools. I am able to learn ~5x faster by simply tapping into asking questions. I can craft solutions, research problems, and develop systems almost 10x and 15x faster, up to 20x on some tasks. I could tell that, even if this takes time, I am freer with one idea.

Does it sound like a dream come true? Well, keep reading.

The requirements for this to happen

The recipe for speeding up my work did not come overnight; it is a constant urge. I run a small company, and any clutter could potentially kill us in the long run. Additionally, I always stand by my name - from small tasks to complex projects, I deal with clients, and they rely on me. To the enthusiasts of Personal Brand - this always has a lot of weight, folks!

The above requires from ITWS and in particular me to be always lean, as much optimized as possible - from processes, documentation, codebase and stacks, hosting, security, even contracting and invoicing if you want. And of course, communication. A lot of hats, but I am still loving to do it. Speed, Quality, and Price are clichés, but we rely on them to shine above other agencies. As absurd as it sounds, even if I pass away suddenly, there is a family and business transferrence plan in place.

You already imagine that it is an essential credo that naturally leads to using the AI in the best possible manner, so:

1. I have started very early. Yes, this is the first good thing to have in place, but it is not vital. You can catch up quickly.
2. I am never shy about software or model spending. One thing I adopted early in 2012 from one of my bosses and leaders, Greg, was to use all software, tools, and paid services until you no longer need them, then cancel immediately. This aligns with Matt's article, as I hear from many people around me that they use the free version. I always use privacy-enabled, no model training options - no ChatGPT at all, primarily Geminy for my paid Workplace plan and a lot of APIs. And I pay for these because they are super cheap rocker boosters.
3. Use in real case scenarios. This is actually important. Luckily, over the last 5-6 years, I have been blessed to work with technologies that are super interesting to me, and it feels like stacking Lego blocks. One piece of knowledge leads to another. But in order to adopt AI, you need to actually integrate it into your work consciously, not just to make a few experiments and decide that from now on, you're done with AI, or that AI will never be as good as you. LLMs were pretty lame in 2023 and 2024, but they caught up, and now in 2026, they are really accurate across a plethora of areas, and this will not get worse.
4. Don't choke on technology. Improve the process, not the model. In 2026, a new model is spawned or updated weekly. War is on, but this does not mean you shall be drawn into solutions, as this is exactly the AI FOMO I told you about. A bit later on this, but the main requirement is: pick a stack; if it does you a good job, try to keep it detached from a vendor lock perspective (use Open Code instead of Claude Code if you can, and it works for you). Try models, but they need to fit and be evaluated against your goals, not just because someone wrote an article about it on Medium, Substack, or X.
5. Data structure and organization. I remember back in 2018, someone told me, "I want the robots to work and replace the humans I have to manage now." It never worked out, as the process turned out to be individualized, chaotic, and strongly personal on so many levels, looking more like a beautiful orchestration of chaos. You need very strict data, an aligned process, and repetitive tasks that do not drift to have successful AI implementation and automations in place. Make these with personal decision in the moment, and AI is inapplicable.
6. A base and fundamental knowledge and experience in your area. This is the most underrated yet most vital requirement. E.g., I have project management, system architecture, technology stacks, databases, middleware, design, content, security, and scalability concerns and considerations for every project I take on, and I flex these skills daily. I see all these comments: "today everyone can build anything with AI," and these are as inaccurate as they are funny. The truth is- "no, you can't" or at least not the way you imagine you will. This is a very large topic in its own, and I will write about it soon. Briefly, if you don't have experience in an area but want to develop there with AI, this is what will happen: the model will build it arbitrarily, piling wrong prompts will accumulate in the context, diluting the end-goal, resulting in something that might look cool, but is far from any industry standard. The website or app looks easy enough, and we already have enough horror stories about vibe coders to know this does not work. Now try this with building an IoT product, a mechanical component, or a specific application that relies on a specific dataset. See, the AI is not a replacement, but an enabler, and as a complex tool, it comes with a plethora of potential pitfalls. You can't debug or help the AI if you don't know what you're doing. This directly applies to all other requirements above.

The disadvantages of heavy AI / LLM usage

1. Token limits hit - I always hit hard limits, even on Max x20, simply because I want to fully use what I am paying for, and I have a plethora of ideas to work on. This makes things scalable at the production level, but it's always dependent on the model. Same with some models, just stop working at some point. If you are completely dependent on them - at the moment your models stop for whatever reason - your process stops too. The deeper the integration, the larger the impact.
2. Models change, sometimes degrade - Mostly for good, and they improve indeed, but there are some specific cases that are really annoying, and the models drift away.
3. Models work differently across stacks and tasks - I often run into older content or inconsistencies (before Claude Code and the general .md guardrails that are now cross-applicable). And some stacks, like React/Next/Tailwind/Supabase, are more popular than others. This is a disadvantage in the general scope, as it will push more lame/less-educated users toward these stacks, and these have a loooot of disadvantages - e.g., far from perfect for any cause. So, you really need to ensure the workflow and the AI you use align with the specific tools (MCPs) you use, and with agents if your process is specific. Example - if you want to run a BDR researcher that collects information for companies on a larger scale, you might need a data scraper, MCPs to connect to LinkedIn or other data gathering tools, then a reviewer of the information, then another one to refine, fill all gaps, and sign off the profile that shall be guaranteed. This shall be wrapped in a translator, especially if you cover markets with non-English-speaking people.
4. Cognitive overload and code fatigue - As fast as development was with Agentic enterprise coding practices, I still needed to audit the code and give the final go. This means I need to review thousands of lines of code, especially if you want to ensure you have a solid solution in place. This is the most exhausting part, as you need to ask as an overseer rather than write the code yourself. If you work on a single product, this can be challenging; if you work across different stacks and multiple solutions, it can be energy-draining. You can't do this constantly for sure.
5. Constant noise, distraction, and FOMO - Like we said, the FOMO that you are missing constantly, something is so widespread that it is almost mirroring the 2021 Crypto goldrush. Shiny new models, new CLIs, endless battlefield. This takes its toll in distractions, experiments, and pointless expenses, until you figure out something that is really working, only to get distracted again. This penalizes the perks and the productivity pretty hard.
6. Security - this is very overlooked, and there are many victims across many industries already. Yed, Open Claw is on the rise 🫠. In ITWS, an example of AI work is done in isolated Docker environments and on feature branches to provide a clear overview and traceability of all work. Logging everything, limiting LLMs, and enforcing safety rules are so essential yet so overlooked by 80% of users. No better time to start in the Cybersecurity sector, yay!

I can go on with more risks, but in general, the picture is clear - this was the reason for my post image to be a fried out Bob, with the laughing man at the center - you can't keep up, and you shouldn't be distracted from what you're doing simply because there is a new shiny thing. Focus on your data alignment, process clarity, and a smooth implementation of tools, rather than cutting through the chaos with an AI torch. Don't jump from one solution to another without testing it in a real-world scenario. Take time to adapt - this happens to everyone - they also need time to adjust daily, and the world is not full of folks who are half cyborgs. You're not late.

Ironically, very few people realize that a tiny fraction of the global population is entangled in this algorithmic bubble, carefully managed by LinkedIn, TikTok, X, and other social networks. Yes, this is completely fake and inflated, and at the moment you stop or delete your data (e.g., TikTok), you stop seeing information for AI. Make a couple of searches, like a couple of posts, and you're again bombarded by information about how the world will end, and that if you don't take AI very seriously, it will take over your life.

Here are some to-go thoughts:

• The businesses will not start to make software for themselves (as they will require maintenance, and we revert back to tribal knowledge, competitive espionage, and data vulnerability)
• The most valuable asset will be the face and reputation of a proven tech professional, not a random vibe-coder.
• Industrial knowledge is a river that will be blocked in the next few years, and this will lead to many unexpected problems, from staff hiring to preserving corporate excellence.
• SaaS, B2C, and other software will enter into a bloodbath. Much more important will be to show integrity, brand, and long-term dedication, since the battle for market share, where you can spawn a twin in a week, will be merciless. This is not necessarily a bad thing, as the winners will be the apps and software that have tapped into previously untapped markets first and have delivered real value to people.
• In regard to the above, I see a need to thrive in areas such as NGO support for underdeveloped countries, the development of reliable solutions, improvements in research and development, and science and well-being. From the other site, the analysis and people profiling will become the new standard, pushing us, whether we want it or not, towards anti-utopia a la Black Mirror.
• We will face entirely new types of data, overload, and burnout. Centers for digital detox, mindfulness, and technology detachment will form, evolve, and attract many young people.
• Young people will become immune to or indifferent toward AI. I see these trends among 10-15-year-olds. And I expect this to transfer to when they enter the workforce.
• Art, music, and writing made and proven to be made by humans will become increasingly valued.
• There will be no AGI soon; this is complete nonsense. AI and LLMs will be limited at some point as we reach the limits of what the systems can support, unless other options for energy and water recycling are found.
• The economy cannot and will not afford to cut too many people off from their jobs, as AI needs tribal knowledge.
• The economy cannot and will not afford to cut too many people off their jobs, as people are the fuel of this very same economy - without paying customers, AI, and the economy will collapse, and implementing a base income will not solve this, as people will still not have enough money to drive the economy ahead.

People will adapt. We always adapt. That is how we are made. So don't lose hope and don't get frightened.