The Changelog: Software Development, Open Source - The overlooked power of URLs (News)

Episode Date: November 3, 2025

Ahmad Alfy explains how URLs are state containers, Shrivu Shankar shares how he uses every Claude Code feature, Yusuf Aytas laments how AI broke technical interviews, Wu Xiaoyun tells how he saved Tik...Tok $300k during his internship, and TOON is a new serialization format to save us some LLM tokens.

Transcript
Discussion (0)
Starting point is 00:00:00 What up, nerds? I'm Jared, and this is ChangeLog News for the week of Monday, November 3rd, 2025. We're making a quick trip to San Francisco next week to partake in sync confit, a boutique conference on the future of real-time, collaborative, and agentic software dev, organized by Johannes Shickling from Prisma,
Starting point is 00:00:28 Adam Wiggins from Heroku, Emma Tracy from Colt Repo, and more. If you're going, let's sync up. See what I did there. If not, stay tuned for the best convos from the hallway track. Okay, let's get in to this week's news. The overlooked power of URLs. Ahmad Alfie found an old comment in his code that contained a powerful link. Quote, I clicked the URL, and it was the PrismJS download page with every checkbox
Starting point is 00:00:58 drop down an option pre-selected to match my exact configuration. Themes chosen, languages selected, plugins enabled, everything perfectly reconstructed from that single URL. Here was a URL doing far more than just pointing to a page. It was storing state, encoding intent, and making my entire setup shareable and recoverable. No database, no cookies, no local storage, just a URL. end quote. URLs can do so much, but we don't always use them to their full potential. In this article, Akhmad explains how URLs are even more than user interface. They're state containers.
Starting point is 00:01:38 They have their limitations, yes, but Pareto tells me we're not benefiting from the virtues of the URL nearly enough. Akhmad agrees. Quote, we've built increasingly sophisticated state management libraries like Redux, MobEx, Zoostand, Recoil, and others. They all have their place, but sometimes. the best solution is the one that's been there all along. How I use every clod code feature. The more I use clod code, the more I want to use clod code. That's a strong indicator of good product design.
Starting point is 00:02:11 The challenge I have is the surface area of the product feels overwhelming. Not that I'm holding it wrong, necessarily, but that I could be holding it better. But if I'm being honest, which I actually always try to be, so I don't know why I feel compelled to prefix the following or any statement, with that phrase, but here we are, aren't we? I don't know how much of Claude Code's feature set is worth investing in. Are slash commands here to stay? Are subagents even worth it?
Starting point is 00:02:37 Will I switch to AMP or Codex or Gemini CLI next week and make any ClaudeCode-specific learnings moot? With those questions, in my mind, I love posts like this one, linked in the newsletter from Srivu Shankar, a brain dump of all the ways he's been using ClaudeCode for me to cherry pick from. More people write posts like this, and I'll keep linking them up for the rest of us. AI broke interviews.
Starting point is 00:03:00 I'm not sure the software industry's interview process was functional prior to October of 22, but as Yusuf Atis laments in this post, it certainly busted now. Quote, everyone now has access to perfect code, perfect explanations, perfect system design diagrams, and even perfect behavioral answers. You don't need a network, you don't need experience, you just need a second monitor, lying. You don't even need that. Check this out, end quote. The this that he referenced in that quote is an interview coder service that has been
Starting point is 00:03:32 recently upgraded with audio support and 20 plus cutting edge undetectability features to keep you invisible across every interview check. It's now time for sponsored news. Postgres for agents is here. Tiger Data just launched Agentic Postgres, the first database built from the ground up for AI agents. We've seen Postgres extended. in every direction. Time series, vector, graph, but this is the next evolution. Traditional databases
Starting point is 00:04:02 wait for humans to query them. Agentic Postgres is designed for autonomous agents that read, write, and reason about data on their own. Its Postgres reimagined for the agent era with built-in memory, context management, and safety controls so agents can collaborate without stepping on each other's data. If you've ever been wondering what the, quote, database for agents looks like, Tiger Data has just answered that question. Learn more at tigerdata.com or check the link in the newsletter to read the blog post. Intern saves TikTok $300,000 per year because Rust. During his internship at TikTok, Wu Zayun ported a core payment service from Go to Rust. Quote, we faced a classic engineering dilemma. How do you squeeze more performance out of a critical system without compromising
Starting point is 00:04:50 stability or breaking the bank. This is the story of how I tackle that challenge by selectively rewriting a performance bottleneck in Rust, resulting in a 2x performance gain and nearly $300,000 in projected annual savings in cloud costs. End quote. One of our industry's favorite principles is premature optimization is the root of all evil, but it's important to note how much heavy lifting the word premature is doing in that axiom. Well-timed optimizations can yield huge wins. like the one Wu and his colleagues deployed. Oh, and if you think this experience soured Wu on Go? Paradoxically, this project gave me an even deeper appreciation for GoLang.
Starting point is 00:05:32 Go's incredible developer productivity and well-rounded performance makes it the ideal choice for 95% of our services. JSON for LLM Promes at half the tokens. Tune, which stands for token-oriented object notation, is a compact, human-readable serialization form, designed for passing structured data to large language models with significantly reduced token usage. The idea here is to still use JSON programmatically, but convert to Tune for LLM input, saving about 40 to 60% of tokens. Why? Because those tokens still cost money and anything that costs developers money will be optimized away as much as we can muster.
Starting point is 00:06:14 That's the news for now, but go and subscribe to the change log newsletter for the full scoop of links. worth clicking on such as the internet archives way back machine link fixer the only thing that matters and if you don't tinker you don't have taste get in on the newsletter at changelog dot news last week on the change log adam jacob joined us to discuss how agentic systems for building and managing infra have fundamentally altered how he thinks about everything including the last six years of his life. And Adam and I finally pushed record on a spooky friends episode all about software projects that are dying, dead or undead. And coming up, on Wednesday, Andrew Nesbitt tells us all about ecosystems, the world's most comprehensive and accurate data set about open source production
Starting point is 00:07:04 and use. And on Friday, we play what should be the most competitive round of our pound-defined game show, because every participant, except Adam, is already a champion. Have a great week. Like, subscribe, five-star review us if you like our work, and I'll talk to you again real soon.

There aren't comments yet for this episode. Click on any sentence in the transcript to leave a comment.