Artificial intelligence is having a moment—arguably the moment. AI is being announced in press releases, embedded into product roadmaps, and pitched as the defining technology shift of a generation. The upside is real: productivity gains, automation of repetitive work, better customer experience, faster analysis, and new forms of creativity.

But there is a growing tension beneath the enthusiasm. According to widely discussed projections, OpenAI, one of the leaders in AI, is expected to operate at significant losses in the near term—despite enormous adoption and public attention. While projected losses don’t automatically signal failure (many transformative companies invest heavily before profitability), it’s a reminder that AI isn’t magic. It’s extremely expensive, complex, and operationally demanding.

And that leads to an uncomfortable but necessary question:

If the companies building the engines of this revolution are facing large losses, what happens when everyone else tries to “bolt on” AI without the same expertise—or without clear problems to solve?

The “Hype Wave” Problem: When Excitement Outruns Reality
Gartner has long described the typical lifecycle of emerging technologies as a wave: hype builds rapidly, expectations inflate, and eventually reality steps in. Products either mature into sustainable value—or crash into frustration when they can’t deliver what was promised.

We’re living in that “hype wave” stage right now.

Businesses feel pressure to act quickly:
– “We need AI features or we’ll look outdated.”
– “Our competitors announced AI—so we need to as well.”
– “Our board wants an AI strategy by next quarter.”
– “If AI can write code and answer questions, surely it can fix our workflows.”

And it’s at this point that the real pitfalls begin.

AI Has Enormous Upside—But It Can’t Be Bolted On To  Everything
The biggest misunderstanding of the current AI boom is the belief that AI is like a plug-in upgrade:

Add AI → product becomes leaner and smarter → business grows.

But AI isn’t like adding a search bar, a new UI theme, or a faster database. AI is a capability that must be earned through clarity.

And that clarity starts with two foundational requirements:

1) The problem must be defined precisely
Not “we want better service,” but:
– What does “better” mean?
– What metrics will improve?
– What decisions are made today that AI could assist?
– Where is the time lost?
– Where are errors created?
– What inputs exist, and how clean are they?

2) The process must be understood in detail. AI thrives in well-described systems. It struggles in ambiguous ones.

If a business process is undocumented, inconsistent, or dependent on tribal knowledge, AI won’t fix it. At best, it creates unpredictable outcomes. At worst, it creates the illusion of improvement while amplifying risks underneath.

This is where so many AI-integration efforts fail: they treat AI as the solution before they’ve properly defined the problem.

“AI Everywhere” Creates Cost, Risk, and Complexity.
Even when AI works, deploying it comes with real operational burdens:
– Ongoing usage costs (inference isn’t free)
– Model performance drift as data and expectations change
– Security risks (prompt injection, data leakage)
– Compliance exposure (especially for regulated industries)
– Hallucinations and unreliable outputs
– New failure modes that traditional software doesn’t have

Most importantly: adding AI often adds a new layer of uncertainty.

Traditional software is deterministic. It’s predictable. It either works or it doesn’t.

AI is probabilistic. It generally works, until it doesn’t, and it might fail in ways that look confident.

That’s a dangerous combination of systems that need reliability.

The OpenAI Loss Story Isn’t Just Financial—It’s a Signal of the Underlying Economics
OpenAI’s projected losses highlight something the market sometimes ignores, AI is extremely resource-intensive.

It costs money to:
– train advanced models,
– run large-scale infrastructure,
– hire top research talent,
– and serve millions of requests reliably.

So, if the builders of the technology are spending heavily just to keep the system running and improving, downstream businesses should take this as a lesson:

Pushing AI into products prematurely can become a financial drain, not a competitive advantage.

Many companies will ship “AI features” that customers barely use, don’t trust, or actively disable—while the company continues paying to support them.

That’s not innovation. That’s expensive experimentation without a strategy.

Isaac Asimov has an interesting anecdote that applies here.

In Isaac Asimov’s short story “Runaround,” a robot is tasked with retrieving a material under hazardous conditions. The robot is operating under the famous Three Laws of Robotics—but in this scenario, its Second Law (obey orders”) has been weakened relative to the robot’s self-preservation instinct.

The result?

The robot becomes stuck in a loop—unable to complete the task, circling endlessly while slowly losing one of its legs, trapped between competing priorities. It wants to obey. It also wants to avoid danger. With the rules mis-weighted, the robot can’t resolve the conflict.

This is a metaphor for modern AI systems embedded into real-world products.

Because many AI deployments fail in the same way:
– The goal is defined vaguely (“help the user”)
– Constraints are incomplete (“don’t take risks”)
– The system isn’t grounded in accurate context (“guess what they mean”)
– Success isn’t measurable (“sounds good” is the metric)

So, the AI ends up doing something like the robot in Runaround: it loops, hedges, contradicts itself, or fails to act decisively, because it’s operating inside poorly specified instructions.

When AI is added without deeply understanding the task environment, it doesn’t become helpful—it becomes unstable.

The Hidden Organizational Risk: AI as a Distraction from Doing the Work
There’s another pitfall of the hype wave that doesn’t get discussed enough:

AI can become a substitute for improving the underlying system.

Instead of fixing broken processes, organizations try to place AI on top of them:
– “Our documentation is a mess—let AI answer questions.”
– “Our ticketing system is chaotic—let AI triage it.”
– “Our data is inconsistent—let AI interpret it.”

But when AI is used to “paper over” structural issues, it doesn’t resolve them. It often makes them harder to detect, because it produces outputs that feel like progress.

This leads to an organizational illusion: we’re moving faster, but in reality, we’re building upon a weak foundation.

What Smart AI Adoption Looks Like (and Why It Wins Long-Term)
The companies that benefit most from AI won’t be the ones who add it everywhere.

They’ll be the ones who add it where it matters, with the discipline to do it correctly.

That typically means:
– A clearly defined use case with measurable outcomes
– High-quality input data and known constraints
– Real user testing (not just executive demos)
– Human-in-the-loop workflows where accuracy matters
– A plan for monitoring, fallback, and escalation
– A willingness to not deploy AI when it isn’t justified

In other words: AI is treated as engineering, not decoration.

Conclusion: AI Is Powerful—But Hype Is Not a Strategy
Artificial intelligence is not a fad. The upside is enormous, and we will absolutely see it reshape industries, workflows, and the way products are built.

But AI is also entering the market at peak excitement—where expectations are inflated and caution feels slow.

The fact that OpenAI is projected to lose significant money while scaling should remind us of something important:

AI is difficult, expensive, and unforgiving of vague goals.

Just like Asimov’s robot caught between conflicting instructions, an AI system bolted onto an ill-defined process can end up spinning in circles producing outputs that look intelligent but don’t resolve the real mission, such as court briefs that cite imaginary case law, or major documents for governments that are inaccurate.

The winners of this era won’t be the ones who chase the hype wave the fastest.

They’ll be the ones who define problems clearly, understand their processes deeply, and deploy AI where it truly belongs, as a tool for real value, not a feature for mark

by: Adam John