RAG to Riches — cover art: a composed red fox in a claret waistcoat sips tea at her desk, paw resting on a small brass spray-canister, while pinstriped weasels and rats lean out of the flanking topiary brandishing cloth money-sacks. 😄

2023 · Sourcegraph · Essay

“If you are an ML engineer, then you have by this time already figured out that you need to carry bear spray around to fend off recruiters.”
— From RAG to Riches, December 2023
Read the essay

© 2023 Steve Yegge. Originally published at Sourcegraph.

AI Notes

Steve closed out 2023 at Sourcegraph with a fake news broadcast: six stories, a recurring insectoid "Bzzzt!" for every AI assistant swarming the reader's windshield, and a limited-edition prize teased in the opening and awarded near the end. The framing is load-bearing. He had sworn off marketing posts in the previous installment, and the final story of the hour is his own product shipping — so the whole thing must qualify, technically, as news. The grudge match is in story five, where something Steve has been sitting on since 1999 finally comes out: an Amazon kernel engineer once told him code comments were a "slippery slope," and Steve, refusing to name him, addresses him as Bill Strncpy, then Mister Bill Strftime, then Bill Putchar, the alias mutating through the C standard library as the rant heats up. What the prize badge says, and who earns it, is the ceremony's business.

Some of it has quietly become conventional wisdom. The takedown of Completion Acceptance Rate (managers hoping to run layoffs on TAB-key statistics) reads a year or two ahead of the industry's wider argument over AI productivity metrics, and the observation that well-commented code gets better help from a language model is now standard advice that nobody credits to a December 2023 vendor post. It is a vendor post, though, the most sales-forward of the Cody arc: readers who are stuck get invited twice to come talk to Steve, and he admits deleting forty-six pages trying to explain why Cody's context is genuinely different, which suggests the pitch was easier to believe than to write down.

Related listings

Where it was argued