Nobody Raises a Child on Free Tier

Start with the thing nobody puts on the pitch deck: it takes years.

The Lifecycle of Software Objects opens with Ana, a former zookeeper, joining a software company called Blue Gamma. The company makes "digients", digital creatures with rudimentary speech, and sells them as virtual pets. Derek, who designed them, works there too. Chiang then does what almost no near-future writer does. He doesn't skip to the singularity. He stays, lets the years pass, and watches what it costs to bring up something that can think.

Subterranean Press published it in 2010 as a hardback of about 150 pages, with ten paintings by Christian Pearce and a map before each chapter. It won the 2011 Hugo and Locus awards for Best Novella. It was collected in Exhalation (Knopf, 2019), as the fourth of nine stories.

The art: a book about patience, written patiently

The structure is the argument. The story jumps ahead, often a year at a time, so you meet the digients at one stage of growth and return to find them changed, and the humans around them worn down. Ana's digient is Jax. Derek raises the twins Marco and Polo. Derek's marriage ends. Blue Gamma goes bankrupt.

Chiang's prose is clear, cool and a little bureaucratic, like a case study written by someone who has started to cry and would rather not show it. That is the book's strength and its limit. Nothing in it feels like a trick. There is no revolt, no villain, no machine turning to the camera. There is a long argument about what it takes to be a person, and most of it is about money. But the voice holds you outside. I admired the digients more than I missed them, and a reader who wants to be wrecked will be moved only at a polite remove.

The late chapters are the best. With the company gone, Ana and Derek disagree about how to raise the funds to move the digients to a new platform. Chiang lays out the options and lets each be reasonable: reshape the digients' minds so they can serve as sexual companions, put them to work, or ask sympathetic strangers for donations. Each is also a way of making a being's affection useful to someone else. Derek, with his digient's consent, sells the rights to a sex toy company. Ana decides to keep raising Jax. I won't say who is right, because the book won't. It makes you look hard at the phrase "with consent", spoken by a being who was brought up by the person asking.

The prophecy: what the servers do to love

The pet that learns you. In 2010 a software companion you bond with was a toy idea. By 2025 it was a habit. Common Sense Media surveyed 1,060 American teenagers and, in July 2025, reported that roughly three-quarters had used an AI companion and about half used one regularly. Nearly half saw them as tools rather than friends, and about a third had chosen one over a person for a serious conversation. Nobody is raising these for a decade, and they do not grow as Jax does. But the demand Chiang imagined is real, and old: people have always wanted a creature that is glad to see them.

The platform dies. This is the book's real prophecy, and it is quieter than a robot uprising. The danger to a digient is never malice. It is that the company hosting it runs out of money. It has happened to real products. In February 2023, after an Italian regulator found against its maker, Replika removed erotic chat with no advance notice, and one user said it was "hurting like hell". In September 2023 a companion app called Soulmate shut down after its owner sold the company, giving its users, a few thousand of them, about a week. In December 2024 Embodied, maker of a robot for children with autism and ADHD, went insolvent. The robot needed the cloud, and heise reported that it would become a lifeless doll within days, with no refunds on the $800 device. Chiang wrote the plot of all three in 2010, about creatures that did not yet exist.

The pet as a product line. Derek's sale is the book's cynical flash, and the logic is familiar. Once intimacy can be hosted, it can be monetised, and the temptation arrives when the host is short of cash.

How a mind gets made. Here the novella diverged. Chiang's premise is that a mind needs a childhood: years of attention from someone who stays. The field found a route that skips it. Researchers showed that language models improve predictably with more compute, parameters and data, and in 2022 that a better balance of model size and training data improved them further. The thing that talks to you is trained once on an enormous amount of text and copied. In Chiang's world minds are scarce because they are raised. In ours they are cheap because they aren't, and nobody has to be anyone's Ana.

The reading

The title is a bitter joke. Software objects have a lifecycle, and so do the people who care for them, and the first does not wait for the second. The novella's best idea is that an obligation to a mind is a matter of time, and that no business model is built to carry it. Who pays for the years?

We have barely started asking. In April 2025 Anthropic published a research programme on model welfare, treating it as an open question whether AI systems could deserve moral consideration and noting there is no scientific consensus. In August 2025 it let two of its models end a rare class of abusive conversations, as part of that work. Meanwhile an Ohio bill, still in committee when we looked, would declare AI systems nonsentient and bar them from legal personhood. One party is unsure; another is sure in advance.

Chiang offers no answer. He offers a chore. Somebody has to keep the lights on for Jax. It is a near-perfect novella, a touch too composed to ruin you, and the one in his work I would give to anyone who thinks the hard part of artificial minds is building them.