You're Not Behind – On AI Anxiety, FoMO and the Invisible Oversight Labor

Hand on heart: how many tabs with AI tools, new frameworks or "must-read" threads do you have open right now? Three? Ten? Or did you stop counting a long time ago?

In this post (Post 19) I wrote about the technical flip side of agents: skill atrophy, cognitive debt, prettily packaged bugs, glowing credit cards. That was the case against the code. This post is the case against the feeling.

Because everywhere on LinkedIn and in tech blogs we read the same story: "AI will 10x your productivity." "If you don't master prompt engineering today, you're out tomorrow." And what I experience in conversations with developers, founders and tech enthusiasts often looks nothing like the great liberation. It's a latent, chronic pressure. The feeling that the train is racing past at twice the speed of sound – while you're standing at the tracks with a skateboard.

Two things up front, so the frame is clear:

First: the feeling isn't deceiving you. The stress is real, measurable, and by now it has a name in the research. You're not imagining it, and you're not alone with it.

Second – and this is the actual title of this post: you still aren't behind. The one does not follow from the other. The feeling of being left behind is an emotional reaction to a broken pace – not an objective finding about your abilities. Keeping those two sentences apart is half the battle.

FoMO-AI – the Fear Has a Name

If you feel stressed because you don't master a new tool every week, you're in good company. Psychology now has its own vocabulary for this phenomenon: FoMO-AI – the Fear of Missing Out, applied to artificial intelligence. The compulsive worry about missing an AI trend and becoming obsolete over it. Recent work on the acceptance of AI tools („I Cannot Miss It for the World", 2024) shows exactly this mechanism: people reach for ChatGPT and friends often not out of genuine curiosity at all, but driven by the fear of otherwise losing touch.

This fear also has a face in the industry. Pluralsight's „New Developer Report" — October 2023, based on over 3,000 developers — describes how a considerable share of them see the value of their own hard-earned skills under threat: around 45 %. The accompanying research by Dr. Catherine Hicks coined the term "AI Skill Threat" for it: the concrete worry that one's core knowledge is devalued overnight.

That number is old by now — and it probably understates things. In the same vendor's AI Skills Report 2025 (April 2025, 1,200 decision-makers and practitioners in the US and the UK), concern about one's own obsolescence sits at 74 to 91 %, a 17-point jump over the previous year. The two figures can't be compared directly, though: different sample, different question — "AI Skill Threat" is a constructed measure, obsolescence concern a direct question. And one more thing belongs on the record: Pluralsight sells training. A vendor whose business depends on people considering their skills at risk is not neutral on exactly that question. That doesn't invalidate the numbers — but in a post about the marketing of fear, leaving it out would be dishonest.

And – this is the part that genuinely gives me pause – it doesn't start with the job. A recent preprint (arXiv:2601.10468) describes how even students feel pressured to frantically reorient their entire education around AI – out of fear of a devalued degree, before they even hold it in their hands. The pressure trickles down a generation before those affected ever had a chance to build a foundation.

The bitter joke in all of this: that very fear drives the behavior that fuels it. You consume more, rush from tool to tool, learn everything only halfway – and afterwards feel not more secure but even further behind, because the half-learned things never congeal into real competence.

The Invisible Oversight Labor

There's a big misunderstanding about the AI hype. The promise goes: generative AI takes the work off our hands. What the marketing slides leave out is what the empirical research shows: AI often doesn't reduce the cognitive load, it merely shifts it.

We are less often the creators of code today. We have become inspectors. In the research this is now called oversight labor. We generate results in seconds and then spend hours hunting hallucinations, checking corner cases and straightening out the generated swamp of tech debt. The paper „At What Cost? Software Developers' Well-Being in the Age of GenAI" (Guizani et al., 2026, also in the ACM Digital Library) names exactly that: a new kind of oversight work that massively increases cognitive load instead of lowering it.

This is the emotional continuation of a finding I already described from the technical side in Post 19 – there it was about the verification bottleneck: that honest reviewing doesn't scale with generation speed. Here it's about the price that same bottleneck extracts from you. Merely interacting with and correcting AI output costs immense mental energy („Developers' Experience with Generative AI", arXiv:2607.02337). And that permanent state has consequences: an empirical paper on modeling developer burnout under GenAI adoption shows that generative AI increases job demands and with them the burnout risk – it doesn't lower it.

The reason is paradoxical: because generation is so fast, expectations about our output pace climb into the astronomical. We're supposed to deliver 10x, but we review at human speed. That gap – fast machine, slow human – isn't comfort. It's a permanent strain.

When Fear Poisons the Culture

Up to here this is an individual problem. But it has an interpersonal and a societal level, and those are almost more uncomfortable still.

Interpersonally, something dangerous happens: when the fear of losing your job rises, psychological safety in the team drops. And whoever doesn't feel safe starts hoarding knowledge. A study on ScienceDirect shows the direct connection: AI-induced job insecurity leads to knowledge hiding – employees deliberately conceal their expert knowledge from colleagues in order to make themselves irreplaceable. Of all things, the technology sold as a collaborative lever drives teams apart.

But it isn't only fear. There's a second explanation for the same behavior, and it's considerably more uncomfortable: plain self-interest.

Whoever is the person on the team who genuinely masters the tools holds an edge — and it pays off. In breathing room, in standing, in being hard to replace. If you finish your work in half the time, you can hand the saved time back or keep it; if you explain the how, you hand it back by definition. Sharing the edge means dissolving it. The distance is the leverage.

So as long as an organization doesn't explicitly reward passing knowledge on, hoarding isn't a character flaw but the rational strategy. And this is the part that gives me the most pause: it isn't only the insecure who hoard then, but of all people the confident ones too — precisely the people the organization could learn the most from. Two entirely different motives, the same outcome: the knowledge stays where it was created. Treat this as a fear problem only, and you solve one half while puzzling over the other.

And then there's a third layer that has nothing to do with psychology at all. In the AI Skills Report 2025 mentioned above, 95 % of tech leaders call AI skills critical to job security — while 61 % of tech workers say that using generative AI at their company is seen as "lazy". At the top it's declared a matter of survival; at the bottom it arrives as a suspicion of laziness. Anyone working openly with the tool in that climate is taking a risk — and anyone doing it quietly certainly isn't sharing any of it. That's no longer a psychological problem, it's an enablement problem. An organization that manufactures fear and stigmatizes usage at the same time is guaranteed to get both: pressure and paralysis.

On top of that, work no longer ends when the day does. Research on AI awareness – the constant consciousness that AI is currently plowing up the market – shows that this very awareness damages the ability to detach and sharpens the conflict between work and private life (PMC / Frontiers). You go home, but the thought "I really ought to learn that one tool" comes along.

Societally, finally, a climate emerges in which many people use AI not out of enthusiasm but simply out of the worry of otherwise falling behind – a collective adaptation pressure that reinforces itself. These three levels stack cleanly – but they aren't a cause-and-effect chain, they're three perspectives on the same phenomenon:

Why You Still Aren't Behind

So much for the assessment. Now to making good on the title – and I don't mean this as a consolation prize, but as a sober observation.

Nobody is up to date. Not even the people who look like it on LinkedIn. The colleague who touts a new framework every other day hasn't finished learning the previous three any more than you have. The market produces tools faster than any human can master them – by definition, nobody can be "up to date" anymore. A race that structurally cannot be won cannot be lost either. "Behind" is relative to a leading edge that doesn't exist.

And the productivity leap that's putting you under pressure is barely detectable at the organizational level. McKinsey's „State of AI 2025" reports that 88 % of organizations use AI in at least one business function — but only around 6 % clear the bar where AI contributes more than 5 % of EBIT. Eighty-eight percent usage, six percent measurable impact. That isn't a success story with a few stragglers. That's a gap between adoption and outcome spanning almost the entire economy.

Field reports from large rollouts keep describing the same distribution behind it — a barbell rather than a bell curve: a small group of genuine power users, a middle group using the tools superficially, and a large majority that barely touches them. One example stuck with me: in an organization with an eight-figure annual licensing commitment, 10 % of the people burned 90 % of the tokens („AI Adoption is a Myth"). The dashboard says "rolled out", the organization's pace says "unchanged" — and both are true at the same time.

What that means for your feeling: the yardstick you measure yourself against isn't the average. You're comparing yourself to the loudest fringe of a distribution whose middle sits far, far further back than LinkedIn lets you believe. You see a gap ahead of you — you never even look behind.

And what is being felt as "lagging behind" right now is something we know from every major technology wave. The pressure isn't new either: people adopt the technology in good part because they fear otherwise being left behind – a pattern that reaches all the way to the front pages. The only new thing is the pace at which that pressure is clocked. But the pace is a property of the hype, not a yardstick for your competence.

The decisive fallacy sits in the equation "knowing more tools = being better". That was never true. What makes software good isn't the number of frameworks you clicked through this week, but a deep understanding of systems – architecture, dependencies, edge cases, business logic. Exactly the competence you don't build through frantic tab-switching, but through the slow, uncomfortable penetration of problems. That ability doesn't age on a weekly cycle. It's what remains when this month's tool hype has disappeared again.

And maybe the expectation itself is the mistake. There's a sentence from enterprise practice I haven't been able to shake for weeks: People are not in the market for a tool that helps them get the work done — they just want the work done. The demand that every person in every profession turn into an AI craftsman is in good part an invention of the market selling those tools. For many roles the right answer isn't "finally learn to prompt", it's this: the automation belongs in the background of the systems people already work in. Whoever never becomes a power user there hasn't missed anything — it was simply never the job.

For us in software development the case is admittedly different: here the tool is part of the craft, and refusing it outright wouldn't be a tenable position in the long run. But even then it stays a tool and doesn't become an end in itself. The measure is what you build with it — not how many of them you own.

How I Handle It: Selection Instead of Sprinting

I haven't found a way to switch the pressure off completely – but I have found one that makes it manageable. It consists of deliberate selection instead of running faster.

• I deliberately learn fewer tools, but properly. I have no fixed cadence and no quota for this — I go by gut feeling. If a topic genuinely appeals to me, I try it out. If it doesn't, I let it pass, and without a guilty conscience. OpenClaw, for instance, simply held no interest for me personally: I picked up on the surface that it exists and roughly what it does — I never actually tried it. Not everything trending on LinkedIn or Twitter is relevant to my project. Sometimes a healthy filter is all it takes. Looking away is a skill, not a weakness.
• I build the control into the structure, not into my head. That's the common thread of my Agentic Coding series. Human-in-the-loop checkpoints at the concept level (Post 12), small validated steps (Post 13), deterministic guardrails from types, linters and tests (Post 14). That same structure is also the best medicine against the oversight labor from chapter 2: whoever receives generated output in small, checked portions doesn't have to wave through a thousand lines of raw AI output while exhausted.
• I deliberately maintain my core competence. As already described in Post 19: every now and then I write, debug and optimize a feature completely by hand – not out of nostalgia, but to keep alive the deep understanding that separates me from being a pure review slave. That's my antidote against skill atrophy and against the feeling of doing nothing but correcting after the fact.
• I switch off – and I mean it seriously. If the research shows that AI awareness sabotages detachment, then deliberately switching off isn't a luxury but self-protection. The next tool release will still be there when I sit down at the machine again tomorrow.

What unites these points: they replace racing with direction. I no longer try to be fast enough to catch everything – that's a fight you can only lose. I try to genuinely understand the few things that count.

Conclusion

Let's be honest: the pressure is real, the research backs it up, and I feel it myself. There are days when I close the tabs and still feel left behind. That's part of this moment in technology history, and I don't want to talk it down.

But the feeling is a poor advisor. It measures the pace of the hype, not the value of your work. The fear of being left behind leads to exactly the behavior – frantic consumption of half-knowledge, hoarding knowledge, never switching off – that actually sets you back. And the way out isn't running faster but choosing more deliberately: fewer tools, deeper understanding instead; less breathing-down-your-neck control, more structure instead; less permanent standby, real detachment instead.

There's one objection to my own optimism I take seriously: every release raises not only the capabilities of the tools but also the bar for using them well. If you have to master agent fleets, skill files and context budgets to extract the maximum, a more powerful model helps you little — the distance between "uses it" and "uses it well" tends to grow with every step up rather than shrink. So it may well be that the pressure won't dissolve on its own.

I'm cautiously optimistic anyway – but by now for a different reason than before: not because the tools get simpler, but because the expectation will normalize. Every technology wave had its phase in which mastering the new tool counted as a personality trait, and every one of them left that phase behind. The technical flip side from Post 19 will probably go the same way. Until then, the sentence I put at the top of this post holds, and I mean it as seriously as I rarely mean one: you are not behind. There is no leading edge you could be standing behind. There is only the question of whether you truly understand the right things – and that isn't decided by the next tool, but by you.

How are you doing with the current pace? Do you feel the oversight labor in your day-to-day – or have you found a way to navigate the hype in a healthy way? I'm honestly curious how others deal with it.