Vibe Everything - The Truth Nobody Wants to Hear

A US military intelligence report concluded that a Chinese vessel travelling through the Middle East was carrying components related to a nuclear weapons programme. Military aircraft were already in the air. Armed personnel were preparing to board the ship. The operation was called off only at the last moment, and all of it had been triggered by an AI hallucination. Sources: EFE · La Vanguardia · Trending Topics · The Independent · CNN · The Times of Israel · Marine Insight

The disturbing part is not that AI can be wrong. Humans have always been wrong. It is that machine-generated output can now move through institutions at extraordinary speed, acquire authority, and trigger real-world action before somebody qualified challenges it.

Table of contents

• Vibe Everything Is Already Here
• Software Engineers Gave It a Name
• We Built the Incentives for This. Don’t Blame AI.
  • Looking Successful Matters More Than Doing the Job
  • The Fast Food Effect: We Save, You Pay
  • Hype Makes You the Problem for Asking
  • We Are Sacrificing the People Who Would Know Better
• Judgement Is the Part We Cannot Outsource

Vibe Everything Is Already Here

The ship incident is not an isolated failure. The same pattern is already appearing in very different fields, including law, consulting, healthcare, and much more. The technologies and consequences are different, but the failure mode is strikingly the same: machine-generated output looks plausible enough to enter a real decision process, while verification arrives later than it should (if it arrives at all).

What matters is not the particular mistake made by each system. It is the pattern around it:

• The output looks convincing;
• Somebody without enough expertise decides it appears good enough;
• An organisation gives it authority;
• Verification, if it arrives, comes later, usually after consequences have already unfolded.

These errors have a name: hallucinations. A generative model can produce an answer that is fluent, confident, and completely false at the same time. Its core task is not to verify truth (at least in the case of LLMs); it generates statistically plausible output from patterns learned during training. That is why the word intelligence can create the wrong mental model: the system can imitate reasoning remarkably well without knowing that a statement is true in the human sense or even what it means.

Hallucination is the technical failure. The real-world failure begins when people mistake plausibility for knowledge and stop applying judgement.

Software Engineers Gave It a Name

Software engineers may simply have been among the first professionals to see and suffer this clearly, because software gives us unusually direct feedback. We see what is under the hood and recognize the gap between something that merely looks like it works and something that is actually well engineered (and really works).

I have been using AI intensively for more than a year, for coding, architecture, research, documentation, analysis, reviewing ideas, and challenging my own decisions. It has become one of the most useful tools I have ever used as an engineer.

But I have never used it on autopilot. Generated code can compile, pass tests, and survive a demo while still containing unnecessary abstractions, bad boundaries, duplicated concepts, and complexity nobody needed. Architecture can be even more deceptive: every individual decision can sound reasonable while the complete system is absurdly over-engineered.

Experienced engineers constantly make a judgement that is difficult to encode in a prompt: how much engineering does this particular problem actually deserve? Sometimes the answer is weeks of analysis. Sometimes it is an afternoon. Sometimes it is twenty lines of boring code. Sometimes the right decision is not to build anything at all. This calibration comes strictly from experience.

Software engineering gave the extreme version of this behaviour a name: vibe coding. But once the same pattern starts appearing in other professions, the name becomes too narrow. We are now witnessing Vibe analysis, Vibe research, Vibe strategy, Vibe management, Vibe intelligence, Vibe decision-making, etc.

Or simply: Vibe Everything.

We Built the Incentives for This. Don’t Blame AI.

AI did not teach us to reward appearances, cut corners or push consequences onto somebody else. We had already built organisations that did all three. AI gave us a way to do them faster, more cheaply and at a scale we could not previously afford. AI did not create this problem. It removed the speed limit.

Looking Successful Matters More Than Doing the Job

You can spend a meeting explaining why the customer experience is getting worse while a dashboard behind you says everything is improving. Tickets are closing faster. Costs are down. Targets have been met. The person presenting the numbers has a success story; the person explaining the damage has an attitude problem. Politics offers the same spectacle: a success story from the people in charge, and a different reality for everyone living with the consequences.

That is what happens when we reward the appearance of progress. A convincing presentation travels further than an inconvenient finding. Investors get a growth story, social media gets a victory announcement, and the people doing the work inherit the gap between the promise and reality. Sometimes that gap is deliberate dishonesty. Sometimes everyone has become very good at believing what benefits them.

AI is extraordinary at giving this a professional finish. It can turn weak results into confident narratives, dress up selective numbers as remarkable achievements, and produce polished explanations for decisions nobody has properly questioned. A struggling project becomes a transformation story. A service losing the people who made it useful becomes an efficiency success. We can now make failure look impressive faster than we can fix it.

The Fast Food Effect: We Save, You Pay

The same appetite shapes the work itself: more, faster, cheaper, with an immediate return. Whether the result remains useful after the demo or the sale becomes somebody else’s concern. This is the fast food effect applied to professional work: optimise for what sells immediately and let the medium- and long-term consequences fall outside the transaction.

Producing quickly is useful when the work is worth producing. Ten times more code that nobody can maintain is a liability. Five reports nobody has verified are five opportunities to make a bad decision. Yet the pressure to do more with less makes the people asking those questions look expensive.

Review, testing, independent verification and experience all cost money. Remove them and the saving appears immediately. The failure they would have prevented is still hypothetical, which makes it easy to dismiss in a budget meeting. We cut the reviewer, keep the deadline and congratulate ourselves on becoming more efficient.

We are borrowing quality from the future to improve today’s metrics.

The bill arrives as broken products, security incidents, bad investments or legal disputes. Employees absorb it through rework; customers through wasted time and unreliable services. By then, the person who claimed the saving may have collected the reward and moved on. The cost lands on another team, another budget or another management. That separation helps explain why the same pattern survives repeated crises: the decision can pay off for its author while damaging everyone who inherits it.

Hype Makes You the Problem for Asking

You might expect those consequences to encourage scrutiny. Instead, each technology cycle supplies a reason to hurry: adopt now or fall behind. Vendors sell urgency, competitors announce transformations, and executives need something impressive to announce in return. Being seen to move becomes a professional survival strategy.

In that atmosphere, ordinary questions become awkward:

• What problem does this actually solve?
• Who checks the result, and who answers when it is wrong?
• What knowledge are we losing when we remove the people doing this work?

Ask them and you risk being labelled resistant to change. The people proposing the shortcut no longer have to justify it; you have to justify slowing them down. You are expected to applaud the decision before anyone has demonstrated that it works.

That is how an experiment quietly becomes production, assistance becomes dependency, and a recommendation becomes a decision. The opportunity can be real and the adoption still be reckless. Hype makes it socially and professionally expensive to acknowledge both at once.

We Are Sacrificing the People Who Would Know Better

The damage goes beyond what we deliver. It reaches our ability to deliver anything better next time.

In my years in software, I have watched teams accept defects, fragile systems and unfinished work that should have been challenged. Customers learn the workarounds. Teams learn to live with the failures. Each compromise makes the next one easier to defend, because the previous compromise is now the baseline.

We like to think of ourselves as critical thinkers, but thinking is work. It takes effort to investigate, tolerate uncertainty and discover that our first answer was wrong. A confident response delivered in seconds offers a comfortable escape. We do not have to be trying to avoid responsibility; sometimes we simply cannot be bothered to do the thinking. AI can help us question our assumptions, but it is just as easy to use it to avoid questioning anything. And every time we skip that effort, we also skip the practice through which judgement develops.

Then we compound the loss: we remove human involvement because it costs too much and slows things down. Sometimes experienced people are replaced by people who cannot properly assess the results. Often, the human check disappears altogether: the system produces an answer and acts on it, with nobody left to question either. We make ourselves dependent on a system while dismantling our ability to challenge it.

Treating AI as a person makes that withdrawal easier to justify. We start saying that it understands the situation, knows what matters and can make the decision. Those descriptions let us mistake a system’s capabilities for the judgement of an experienced professional. Once we believe somebody competent is already handling the problem, human involvement starts to look redundant. We stop asking who is checking because we think we know who is thinking.

Return to the ship at the beginning of this article. Experienced analysts recognised that the claim was false. Now imagine treating that expertise as the next cost to eliminate. Who stops the operation then?

Judgement Is the Part We Cannot Outsource

Nobody seemed confused about who deserved the credit when the savings were announced. They had made the organisation faster, cheaper and, according to their own presentations, far more effective. The transformation was their achievement. They collected the praise for their vision and leadership.

Then something went wrong, and suddenly AI had made the decision. AI misunderstood. AI got it wrong. How convenient: the people who were so visible when there was credit to collect disappear when there are consequences to explain.

We know this routine. Long before AI, success belonged to whoever presented it, while failure belonged to another team, a supplier or an employee who had apparently misunderstood the instructions. AI gives that old habit an exceptionally convenient scapegoat: one that cannot object, challenge our account or remind anyone who approved the decision. Taking the credit and passing the blame has never required much imagination. Now it requires even less effort.

The report still carries our name. The decision still has consequences for somebody else. Accountability cannot be outsourced to a probabilistic model.

And this does not stop with executives. Every time we submit work we have not checked or approve an explanation we cannot defend, we ask somebody else to trust a judgement we never made. We are very good at recognising other people’s mediocrity and finding practical reasons for our own.

Keeping a human in the process means little if their only acceptable response is approval. They need the knowledge to recognise a problem, the time to investigate it and the authority to stop it. Otherwise, we have kept someone to take the blame while removing their ability to prevent the failure.

The temptation is simple: let the system do the thinking and call our willingness to accept the answer efficiency.

AI does not replace judgement. It exposes where judgement no longer exists.