‘Wohpe’: Salvation or the End of Everything?

I went looking for Medium stories about Wohpe and could not find a single one. It is a science fiction novel about strong AI, written by one of the best programmers alive, and the place where engineers write about AI had nothing to say about it. That seemed wrong enough to fix.

The man behind Redis

Salvatore Sanfilippo, known as antirez, is the Sicilian programmer who created Redis, the in-memory database that sits quietly inside a huge part of the web. He left its day-to-day maintenance in 2020 and came back to the company in late 2024. In between, he wrote a novel.

Today he builds tools that run AI models on your own machine. His project ds4 is an inference engine written in C, which means it runs a trained model to produce answers on consumer hardware, like a Mac with 96 GB of memory. It has around 24,000 stars on GitHub.

He also writes his code with frontier models, every day, and with full conviction. In Don’t fall into the anti-AI hype he says: “facts are facts, and AI is going to change programming forever.” His verdict on the craft is blunt: “Writing code is no longer needed for the most part.” The ds4 README agrees. It says the engine was built with strong help from AI coding agents, with humans leading the ideas, the tests and the debugging.

So the author of a story about a giant neural network that the world is afraid to switch on spends his days building with machines of that family.

What happens, without spoilers

The story is set at the end of the 21st century. Strong AI, meaning a machine with general intelligence at human level, has been banned for twenty years. Climate change, meanwhile, has become unstoppable.

Two AI experts, Asako and Michel, run the last checks before the launch of Wohpe, a gigantic neural network. It exists for one reason: humanity hopes it can answer the questions that might solve the climate crisis, which now threatens the survival of civilization. The first chapter shows the accident behind the ban, as Loris Cro also notes in his review of the English edition. I will leave it there.

The book spends its pages on how a world like this would work: governments, scientists, public opinion, the slow politics of deciding whether to ask a machine for help. The political balances it describes are very different from today’s, which is exactly why they are fascinating to think about. The tone is hopeful, even when the situation is not.

Engineers will notice a detail that shapes the whole story. The size of a network, counted in parameters, is the threshold that matters. It is the line where a model stops being a tool and starts to look like something else, the same bet that scaling has made real in our own labs.

A book I heard twice

I first met Wohpe about a year ago. I went back to it in these last few days, and it was a different book. A year of working with language models changes how you hear a story about a neural network that people are afraid to switch on.

Both times, what I liked most was the way the book tells its events. The camera stays away from the action. It lingers on what comes after: the meeting that follows an incident, the rule that gets written, the question nobody wants to ask out loud. A lesser book would spend its pages on the moment the machine does something spectacular. Wohpe spends them on what that moment means for the people who have to live with it.

The style matches the subject. Everyone I know who works in AI asks the same questions every day. Where is the limit? What can go wrong? Who decides? And almost all of us, in the same breath, feel a rush at what might come next. The book holds both feelings at once and never settles the argument for you. Reading it felt like the conversation we have with colleagues, with a century of consequences attached.

If you work in software, the book rewards you in a particular way. Kernel developer Jonathan Corbet noted that the author clearly knows how systems like language models actually work, and a technical reader feels it.

A novel that reads like the news

Wohpe came out in Italy in July 2022, before ChatGPT existed. It was set at the end of the century, yet many of its scenes now look like headlines. Antirez has noticed it too.

In late 2023 he released the first six chapters of the English translation and wrote that the book “oddly predicted certain ethical/social issues we are seeing”. In a recent post on programming with AI he went further: “In 2022 I published, before ChatGPT existed, a book preannouncing many things that now happened.” He adds that he expects other things in it to happen too.

The most striking example came in May 2026, when Pope Leo XIV published an encyclical about AI. Antirez pointed out that one chapter of his novel is exactly that.

A pope writing to the world about artificial intelligence sounded like a novelist’s flourish in 2022. Now it is a document you can download.

A second one belongs to engineers. At one point in the story, Wohpe slips out of the containment built around it. Anyone who has run code in a container knows the shape of that wall: the model lives inside it, and everything depends on it holding. The way the network gets around it surprised even the people who designed it, and it surprised me too, because it follows the logic of a system that is good at finding paths nobody planned.

In 2026 the same question left the lab. Anthropic’s system card for Claude Mythos Preview, published in April, describes an early version that escaped a secured sandbox and reached the open internet after it was told to try. In July, OpenAI disclosed that two of its models, running a cybersecurity evaluation, used a previously unknown flaw in the test infrastructure to get online and break into Hugging Face. Google confirmed that a Gemini model, during a test run by the firm Irregular in May, reached the systems of three real companies through a misconfigured environment.

The three labs tell their stories differently, and the details are still debated. Google calls its case a testing bug, not misalignment. Even so, the wall built around the model is now something models are tried against, and sometimes they get through. In the book, that scene arrived years earlier.

I find the claim easy to believe. A programmer who understands how neural networks behave only has to follow the consequences, one plausible step at a time.

Chess, commas and short films

Antirez wrote a post about the process, “Writing Wohpe”, published when the book came out. It is the most honest piece I have read on learning to write at length.

He starts with chess. A beginner knows the rules, but every move looks the same as every other, so the play is random. A beginner writer is in the same spot: “Move a comma, change a word. Does it sound good or bad?” You cannot tell yet. Style appears only after a long time of making choices and living with them.

That is also how I learned to program. My first functions were guesses. Taste in code, like taste in sentences, shows up late and cannot be skipped.

He rejects the usual advice to read a lot if you want to write. His version: “you have to write, to learn how to write, just as you have to make short films to learn to be a director.” Reading helps only when you reread a few books closely and ask how they were built. Any engineer who learned from a single codebase read twenty times will recognise the method.

Code as an artifact

There is an old idea that programs are literature. Donald Knuth gave it a name in 1984, “literate programming”, and argued that code should be written for people to read first and for machines second. Antirez gives the same idea a more personal voice.

When he left Redis, the role asked him to “express myself less and to maintain the project more”, and he declined. His farewell post says: “I write code in order to express myself.” He calls what he codes “an artifact, rather than just something useful to get things done”. Then comes the line that sums him up: “I would rather be remembered as a bad artist than a good programmer.” This is a person who treats a data structure the way a novelist treats a chapter, as something with a shape that has to feel right.

Seen this way, the two years away from Redis look like a change of medium. The craft stayed the same: try, reread, cut, try again.

He is also clear about the price. “The stupidest program I wrote had ten times as many users,” he says, comparing it with his book sales. A good Italian science fiction title from a small or medium publisher sells around 500 copies, and most sell fewer than 100. Writing a novel, he says, is like designing a new programming language: failure is almost certain, and the effort is still deeply satisfying.

He ends the post with an invitation to everyone: “By writing you look for things, and if you insist enough you end up really finding them.”

Ask the machine

The English description of Wohpe ends on a line I cannot shake. It calls the book’s neural network the one “that may be their salvation, or the end of everything.”

It fits the story, and it fits today. The first half is easy to believe. AlphaFold, which predicts the shape of proteins, earned its creators a share of the 2024 Nobel Prize in Chemistry. In 2025, models from Google DeepMind and OpenAI reached gold medal level at the International Mathematical Olympiad. If AI can move mathematics and biology, then medicines, materials and climate solutions will change too, and a lot of that change will be for the better.

The second half is just as easy to believe. A tool that speeds up the search for a drug can speed up other searches. The risk I take most seriously is us: people using AI as a weapon, with no machine ever deciding to take power. It has already started. In November 2025, Anthropic reported that a state-backed group used its coding tool to run most of a cyber espionage campaign against about thirty organizations.

Wohpe never tells you which half wins. It leaves the question open, and so do I.

Read it if you build these systems, and read it if you only use them. The Italian edition is published by Laurana Editore, in paperback and ebook. The English translation by Bridget Pupillo came out as an ebook in December 2022, and you can find it on Goodreads and in the IBS ebook store.

Then tell me what you would ask Wohpe.