“Put AI in everything” has become almost a reflex. But not every reflex is rational.

We are entering a phase that strongly resembles something engineering already knows well: overengineering. Only now, applied to artificial intelligence.

I call this OverAI.

What OverAI is

OverAI is not about using AI.

It is about the excessive, performative and indiscriminate use of LLMs and GenAI - often without technical need, without clear return and, most importantly, without sustainability.

It is not a critique of AI. It is a critique of how we are adopting AI.

The parallel with overengineering

The analogy is not aesthetic. It is structural.

Overengineering:

  • using complex architecture without need
  • adding abstractions without return
  • creating systems larger than the problem

OverAI:

  • adding LLMs where deterministic software would solve it
  • using agents where simple automation is enough
  • turning AI into an aesthetic requirement
  • measuring adoption instead of outcome

A classic inversion takes place here:

technology stops being the means and becomes the end.

Tokenmaxxing: the caricature of the problem

The tokenmaxxing phenomenon exposes this almost didactically.

Companies have started to:

  • create internal rankings based on token usage
  • treat consumption as a proxy for productivity
  • use AI as an organizational status marker

This is a direct case of Goodhart’s Law:

“When a measure becomes a target, it ceases to be a good measure.”

Tokens measure computational consumption. Tokens do not measure:

  • quality
  • impact
  • clarity
  • generated value
  • business outcome

It is a classic confusion between input and output.

The economic incentive behind it

This behavior does not emerge by accident.

The AI industry has clear incentives to increase consumption, increase dependency and increase computational throughput.

The more prompts, agents, context, inference and automation, the greater:

  • GPU consumption
  • cloud revenue
  • token consumption

There is a market force pushing the narrative that:

more AI = better.

Even when that does not yet hold economically.

AI as cultural identity

The problem is not just technical. It is cultural.

AI has stopped being just a tool and became:

  • a status symbol
  • a marker of belonging
  • a technological aesthetic
  • social performance

This is not new. We have seen it before with hustle culture, biohacking and startup grind.

Now we have:

AI grind culture.

OverAI as a phase of the hype cycle

Every technology goes through a predictable cycle:

  • discovery
  • hype
  • irrational adoption
  • frustration
  • maturity

We have seen this with the internet, microservices, blockchain and Kubernetes. AI is following the same pattern.

It is not about “popping the bubble”

It is more subtle than that.

The point is not that AI will “fail”. It is that the market is overestimating where it actually creates value.

The most likely outcome is:

  • consolidation
  • reduction of excesses
  • abandonment of bad use cases
  • persistence of the ones that pay for themselves

Exactly as it happened with other technological waves.

The organizational problem

The most critical point is perhaps this:

we are starting to measure proximity to AI, not value delivered.

This produces operational theatre, empty metrics, performative consumption and artificial complexity. And engineering suffers - a lot - when this happens.

Not everything is OverAI

It is important not to fall into the opposite extreme.

There are areas where LLMs have already proven structural value:

  • copilots
  • semantic search
  • document automation
  • operational support
  • data analysis
  • coding assistance

The problem is not AI. It is the disproportionate and indiscriminate use of it.

Conclusion

OverAI is not about rejecting AI. It is about refusing irrational adoption - and perhaps the clearest signal of that is simple:

when we start measuring consumption instead of outcome.

AI will continue. The excess will not.