Weeks of Work Done in 20 Minutes: Redis Creator Says the Era of Hand-Writing Code Is Over

The author of this article, antirez, is the creator of the famous in-memory database Redis. His software powers the operations of tens of millions of businesses worldwide. antirez was once an AI skeptic -- he even wrote a novel expressing his fears about AI automation replacing humans. But his recent experience has proven that the reality of programming has changed forever. In antirez's tests, Claude Code demonstrated epoch-defining efficiency. It can write 700 lines of rigorous, high-performance C code in 5 minutes. It can autonomously iterate to reproduce and fix extremely tricky deadlocks and timing bugs in Redis. Even complex internal restructuring of Redis Streams took only 20 minutes. antirez believes programmers today should stop fixating on typing out code line by line, because in most cases it is no longer the rational choice. He suggests everyone learn to become a great collaborator with AI, channeling energy into defining problems and designing solutions. He encourages every developer to start learning and deeply using Claude Code. As antirez puts it: do not miss your chance to shape the future by refusing to face reality. Spend a few weeks testing these new tools and find ways to multiply your productivity. The purest joy of programming is building, and AI will let you build more, build better, and build with more joy.

I love writing software, one line at a time. You could say my entire career has been a pursuit of writing clean, elegant, tasteful code. Honestly, I did not want AI to win economically. I do not care whether the current economic system gets disrupted by AI -- if it leads to massive wealth redistribution, I would even be happy about it. But deliberately pretending reality does not exist because of my views on software and society would be a betrayal of my own reason. Facts are facts. AI is changing, and will permanently change, computer programming.

In 2020, I quit my job to write a novel. It was about AI, universal basic income, and a society adapting to labor automation amid various challenges. In late 2024, I started a channel dedicated to discussing AI, how it is used in programming tasks, and the possible impacts it may bring. Although I recognized early on that these changes would happen, I always assumed we had at least a few more years before programming would be truly reshaped. I no longer believe that. Today, with state-of-the-art large language models, you can give a clear set of objectives and have the model complete large sub-tasks -- or even medium-sized projects -- almost entirely on its own, with minimal human assistance. How well it works depends on the kind of programming you do (the more self-contained and text-describable the task, the better -- systems programming is a great fit), and on your ability to think the problem through clearly and communicate it accurately to the model. But overall, it is now clear: aside from doing it purely for fun, hand-writing code in most projects is no longer the most rational choice.

Over the past week, I have mostly been writing prompts, occasionally reviewing code and offering guidance. In just a few hours, I completed the following four things -- work that would have taken me weeks before.

I modified my linenoise library to support UTF-8. I also built a line-editing test framework with a simulated terminal that reports the contents of every character cell. This was something I had always wanted to do. But it was hard to justify investing so much effort into a small sub-project just for testing purposes. Now you just describe what you want, and it becomes code. The situation is completely different.

I fixed intermittent failures in the Redis test suite. This kind of work is extremely tedious, involving timing coordination, TCP deadlocks, and similar issues. Claude Code iterated continuously, spending enough time reproducing the problem, inspecting the state of each process, figuring out what was happening, and then fixing the bug.

Yesterday, I wanted a pure C library capable of running inference with BERT-style embedding models. Claude Code wrote it in 5 minutes. The library's inference efficiency is roughly on par with PyTorch, only about 15% slower. The entire library is just 700 lines of C code. It even generated a Python utility for converting the GTE-small model.

Over the past few weeks, I have been modifying the internals of Redis Streams. I wrote a design document for this work. I handed the document to Claude Code, and it essentially redid all of my work in about 20 minutes. The parts that took longer were my own reviewing and confirming commands.

What is happening now is impossible to ignore. In most cases, humans no longer need to write code themselves. The more interesting part has become figuring out what to do and how to do it. On the "how to do it" front, large models are excellent collaborators. Whether AI companies can recoup their investment, or whether the stock market crashes, does not matter in the long run. Whatever outrageous or absurd things the CEO of some unicorn company says does not matter either. Regardless, programming has been permanently changed.

For me, the fact that code I have written was used to train large models is a wonderful thing. I see it as a continuation of a lifelong pursuit: democratizing code, systems, and knowledge. Large language models will help us write better software, faster. They will also give small teams the ability to compete with large corporations. Just like open source software did in the 1990s.

As a programmer, I want to write open source now more than ever. I want to pick up some repos I had shelved due to lack of time and improve them. I want to go all-in on AI within my own Redis workflow. First improving the Vector Sets implementation, then moving on to other data structures, just as I am doing now with Streams.

But I worry about those who will lose their jobs. It is still unclear exactly what changes will unfold. Companies may want more people to do more things. Or the opposite may happen: fewer but more AI-proficient programmers, to keep salary costs down. I also worry that in many other industries, humans may become almost entirely replaceable.

As for how to program. I have just one piece of advice for you, my friend. Whatever you believe is "the right thing" cannot be influenced by denying reality. Choosing not to use AI does neither you nor your career any favors. Think about it carefully. Try these new tools seriously. Spend a few weeks, not five casual minutes just to confirm what you already believe. Find ways to multiply yourself. If you do not find the right approach at first, try again in a few months.

You might think about how hard you worked to learn programming, and now machines write code for you. But the fire that kept you up late at night, just to see your project come alive -- that was the joy of creation. If you find the right way to use AI effectively, you will create more and create better. The joy is still there, fully intact.