May 4, 2026
Cache me outside, AI edition
Redis array: short story of a long development process
Redis gets a big new feature, and the comments instantly turn into an AI workplace fight
TLDR: Redis’s creator spent four months adding a major new feature, saying AI helped him build something more ambitious but did not replace hands-on human work. Commenters turned that into a bigger debate about AI hype, warning bosses not to treat this as proof they can swap out developers.
The big news is that Redis creator antirez spent four months building a new array feature, with artificial intelligence acting less like a magic robot and more like an intense coding sidekick. He says AI helped him go further, rethink the design, rewrite weak parts, test aggressively, and even add search features after he started stuffing markdown files into the system. In plain English: this was not a quick gimmick. It was a long, careful build by one of the biggest names behind Redis, the ultra-fast tool many apps use to store data.
But the real fireworks are in the comments. One camp basically said, “Yes, this matches my experience exactly: amazing helper, terrible replacement for an actual brain.” Another commenter immediately slammed the brakes for any startup boss dreaming of replacing developers, warning that if the original creator of Redis needed four months and full involvement, this is absolutely not a green light for CEOs to force every engineer onto AI coding tools. That’s the drama: is this a victory lap for AI, or a cautionary tale about overhyping it?
And then there’s the lovable internet chaos. One person wanted the secret sauce spec file and a YouTube explainer, another complained Safari mobile showed basically nothing at all, and someone else wondered if Redis is quietly turning into a “small database.” So yes, the code landed—but the comments section landed a whole second storyline: AI hype, boss panic, product creep, and one very broken mobile page.
Key Points
- •The new Redis Array data type took about four months to develop, with the first month focused on a detailed specification.
- •The author used AI tools including Opus, GPT 5.3, Codex, and GPT 5.x for design, implementation, review, and testing assistance.
- •The array implementation evolved from a two-level sparse design to a super-directory structure with sliced dense directories and default slices of 4,096 elements.
- •The design goal was to support very large sparse indexes efficiently while allowing ARSCAN and ARPOP to run based on existing elements rather than index span.
- •The project expanded to include ARGREP and regex support using the TRE library, which the author optimized and further tested.