The Changelog: Software Development, Open Source - AI makes tech debt more expensive (News)

Episode Date: November 18, 2024

Evan Doyle says AI makes tech debt more expensive, Hunter Ng researches the ghost job ad phenomenon, Gavin Anderegg analyzes Bluesky in light of its recent success, Martin Tournoij rants against best practices & Evan Schwartz tells us why he thinks binary vector embeddings are so cool.

What up nerds, I'm Jared and this is ChangeLog News for the week of Monday, November 18th, 2024. Is the love song dying? Well, thanks to some amazing investigatory and visual work by the folks at Pudding, we now know that modern pop music's treatment of love and romance isn't actually in decline. Or maybe it is. It depends on your definitions of love and romance. Go experience their piece, which is linked up in the newsletter, and decide for yourself. What I learned, I agree more with their hypothetical boomer Bob than I'd like to admit. And I just admitted it. Oh well, let's get into this week's news.

AI makes tech debt more expensive. Evan Doyle says what we've all been feeling. Quote, there is an emerging belief that AI will make tech debt less relevant. Since it's getting easier to write code and easier to clean up code, wouldn't it make sense that the typical company can handle a little more debt? The opposite is true. AI has significantly increased the real cost of carrying tech debt. The key impact to notice is that generative AI dramatically widens the gap in velocity between low debt coding and high debt coding, end quote. His advice, instead of trying to force CodeGen AI tools to tackle thorny issues in legacy codebases,

rely on humans to refactor said codebases until they are manageable enough for AI to operate on them smoothly. This changes the makeup of modern development teams. Quote, a product should be owned by a lean team of experts focused primarily on the architecture of their code rather than the implementation details. End quote. Evan goes on to describe why it's more valuable than ever to have a high quality code base and invest in modular architecture early on to keep your productivity tools producing. Up to 21% of job ads may be ghost jobs. We've been talking about fake developer job postings around these parts for a while, but now someone's gone and done some actual research on the phenomenon. Quote, using a novel data set from Glassdoor and employing an LLM BERT technique, I found that up to 21% of job ads may be ghost jobs, and this is particularly prevalent in specialized industries and in larger firms.

The trend could be due to the low marginal cost of posting additional job ads and to maintain a pipeline of talents. After adjusting for yearly trends, I find that ghost jobs can explain the recent disconnect in the beverage curve in the past 15 years. The results show that policymakers should be aware of such a practice as it causes significant job fatigue and distorts market signals. End quote. If you don't mind, I'll remain dry by copy pasting from episode 115 when these ghost jobs were first percolating through our community. In that episode, I said this, be careful out there and give yourself a little leeway too. Maybe you didn't get the job, but then again, maybe nobody got the job. Maybe Blue Sky has won. There's been a lot of hubbub about Blue Sky's current moment,

but instead of linking to yet another article on The Verge that states the growth trend, Here's Gavin. or decentralized, I'm here to tell you that it's currently neither. This one really irks me, because the service is getting the credit for work it hasn't done. One problem here is that the whole decentralized thing is complicated. I believe the Blue Sky team is putting in a lot of good faith effort to becoming a decentralized platform, but this work is tricky because their architectural choices are quite novel, end quote. The at protocol is certainly interesting, but its promise is nowhere near realized, and I'm personally skeptical that it ever truly will be. VC is a hell of a drug. Here's Gavin's big picture takeaway, quote, I don't know what the social media landscape will look like in six months, but I bet things will change. If Blue Sky comes out as

a winner and more posting happens there, I think I'm generally fine with that, at least for now. The whole Twitter mess has taught me not to attach myself too closely with these things anymore. I hung on far too long to Twitter while it made me feel terrible.

End quote. If you listened to last week's friends, you already know my current big picture stance on social networks, but if you are on Blue Sky, connect with changelog.com for sure. It's now time for sponsored news. Sentry launch week starts today. Join our friends at Sentry for daily video drops on YouTube and Twitter starting today at 9 a.m. PST.

You can also hop on Discord to chat live with the engineers building Sentry. Here's four days of new features that you probably won't hate. Day one, smarter search, uptime monitoring, and updates to session replay. Day two, AI-powered issue grouping, autofix, anomaly detection, and more. Day three, domainain-Specific Application Performance Insights and Continuous Profiling. Day 4, Automatic Unit Test Generation, Flaky Test Detection, and AI PR Review. Follow the link in your chapter data and the newsletter to read more about it and get notified about the updates. And thank you to Sentry for sponsoring Changelog News. Against Best Practices to Sentry for sponsoring ChangeLog News. Against best practices. Here's Martin Tornolj.

Tornij. Tornij. T-O-U-R-N-O-I-J. Quote, I have come to believe that by and large, best practices are doing more harm than good, not necessarily because they're bad advice as such, but because they're mostly pounded by either one, various types of zealots, idiots, and a-holes who End quote. The linked piece is certainly ranty, but it strikes as something many of us know, but is easily forgotten. And this is that laws and best practices are generalizations. Generalizations are powerful, but necessarily lacking any specific context. And the hard thing about software engineering is that context is king. This means we often have to eschew best practices when they don't apply in our context. Deciding when to follow

and when to eschew requires thinking and decision making. That's a lot of work. We'd rather just follow the rules and be done with it, wouldn't we? See also Cargo Cult Programming. Binary vector embeddings are so cool. Here's Evan Schwartz, quote, vector embeddings by themselves are pretty neat. Binary quantized vector embeddings are extra impressive. In short, they can retain 95 plus percent retrieval accuracy with 32 times compression and about 25 times retrieval speed up. Let's get into how this works and why it's so crazy. End quote. If you haven't been primed on embeddings yet, they let you turn arbitrary pieces of text into a series of numbers which let you easily search for pieces of content

that have similar meanings by finding the similarity between those numbers represented by vectors. This stuff is at the core of LLMs and modern semantic search techniques. The binary quantized vector embeddings, that's a mouthful, that have Evan super impressed are effectively a lossy compression mechanism for vector embeddings that make them much cheaper and faster to use without losing much accuracy. Quote, That's the news for now, but also scan the companion newsletter for even more news worth your attention, such as CSS's new logo uses Rebecca Purple. And why?

WasmVision gets you going with computer vision, an open source alternative to Google's Notebook LM, and more. Check that out using the link in your show notes or at changelog.com slash news. We have some great episodes coming up this week. Helena Zhang and Toby Freed from Phosphor Icons and the rad new Departure Monophont join us on Wednesday. And are local first apps truly the future? Johannes Schickling and James Long join us to discuss and debate that question on Changelog and Friends on Friday. Have a great week, get yourself some merch while it's still on sale, and I'll talk to you again real soon.