This short post is a pragmatic take on the “AI is hype” backlash from antirez (Redis creator): even if you dislike the incentives, dislike the current crop of AI companies, or worry about the social fallout, the capabilities are now too useful to ignore.
The core claim is that coding has already changed: with today’s top models, many “weeks-long” tasks can collapse into hours if you can clearly specify the outcome and keep a tight feedback loop (running tests, reviewing diffs, and steering the agent when it goes off course). The author lists concrete examples from their own work: adding UTF‑8 support and a new test harness to linenoise, chasing timing-related Redis test flakiness, generating a small C inference library for BERT-style embeddings, and reproducing an internal Redis Streams refactor from a design doc.
What makes the post feel balanced is that it doesn’t pretend this is all upside. It explicitly worries about centralization (a few labs owning the best models) and about job displacement. The practical advice is: treat AI as a tool that multiplies your output, and invest real time learning how to direct it well — while also pushing politically for safety nets and for open models so the benefits don’t concentrate in a handful of companies.