Most coverage treats the recent stories about children grieving robot companions as a touching-but-cautionary tale about emotional attachment. A kid bonds with a device. The device breaks or gets discontinued. The child experiences loss. Cue the think pieces about parenting in the age of AI.

This framing misses the real problem. What we're seeing is not a failure of children's emotional development. It is a signal that we are deploying applications without understanding their systemic consequences.

Consider what happened: A company created a product explicitly designed to be emotionally engaging. It worked. Children formed attachments. Then what? The company faced technical or business pressures and the product died. The children experienced grief. We act surprised.

But this was predictable. More importantly, it reveals something broken about how we are currently building AI applications.

The pattern is everywhere once you look for it. We design AI tools to be maximally helpful and intuitive. They integrate into workflows, relationships, and daily routines. People depend on them. Then a startup pivots, a model gets deprecated, or a company runs out of funding. The application vanishes. Users are left scrambling.

This happens in enterprise settings too, though we hear less about it. A team adopts an AI tool for code generation or content workflows. It becomes integral to their process. Then the API changes, pricing becomes untenable, or the vendor gets acquired and the product gets shelved. The team loses productivity. Knowledge walks out the door.

The robot companion stories matter because they make the invisible visible. Children cannot hide their emotional response to loss the way a corporate productivity team can. So we see the human cost clearly.

But the cost exists regardless of whether we acknowledge it.

Here is what concerns me: We are building applications as if they will always exist. We design them to create dependency. We encourage integration into intimate spaces, both emotional and operational. Then we treat discontinuity as an externality, something that happens to other people's problems.

The recent commentary on AI fragmentation in commercial applications hints at this too. As different systems cannot talk to each other, users get locked into specific platforms. This creates switching costs. When a platform fails, those costs become very real.

We should be designing differently.

Applications should be built with their own mortality in mind. This does not mean they should be poorly designed. It means they should be architected so that when they end, the transition is manageable. It means being honest with users about what we do not know about long-term impacts.

It means resisting the temptation to make every application emotionally engaging simply because we can.

Some of this is cultural. Silicon Valley optimism tells us that better technology solves problems. Sometimes it just creates different ones. A robot companion might genuinely help a lonely child. But if the product model cannot sustain that companionship, then the question is not whether to build it, but how to build something sustainable instead.

Some of this is structural. Companies need business models that do not require constant growth. They need to be willing to maintain products even when they are not expanding revenue. This is not profitable. That is precisely why it matters.

The deeper issue: We are treating AI applications as discrete objects rather than as systems that will shape behavior and create dependencies. Until we think about what happens when those systems fail, we will keep being surprised by the grief that follows.

The robot companion crisis is not a story about children and attachment. It is a story about builders who did not think through what they were building.

That should concern us, because it is happening everywhere else too.