I’m happy to get this series going again after a couple of years away. Today’s list is largely drawn from the projects I’ve been working on for the last few months. As you can imagine, lots of AI and AI adjacent things. There’s also been a lot of boilerplate-y research, because every project needs to start with a day (or more) of yak shaving angst about the “best” language & framework to use.
I’m very curious what everyone reaches for first when setting up a new backend these days. Over the last three years, I’ve used NextJS for frontend and some API backends, NestJS for some pure backends at the startups, and explored Elixir/Phoenix, SvelteKit, and Rust/Rocket. At Disney, we had very smart people devise custom frameworks for our web & API products, which I miss from time-to-time (they were only available internally). Nothing has been as productive to me out of the box as Rails was back in the day, which is sad - I prefer Typescript these days, and am tempted by Elixir and Rust.
(For what it’s worth, I picked Fastify & Apollo for my current project because they looked like they’d get out of the way quickly and I know the Apollo framework well enough. Ultimately, my efficiency is paramount as a solo/small team dev).
Reads #
-
Everything my doctor told me was basically right (or lessons from using Stelo for 3 months): I’ve been focused on improving my overall fitness for the past few months. I summarized one set of changes I’ve put into place along with my use of a continuous glucose monitor to track and measure improvements. These devices are now available over the counter and can be a helpful tool for improving fitness and, for more serious athletes, workout and training efficiency. You can read more about my journey (both into this mess and how I’m starting to get out of it) in this post on my personal (non-tech) blog.
-
Putting the New M4 Macs to the Test: The new M4 MacBook Pros are wowing a lot of folks and their benchmarks. For AI apps, this blog entry by Roboflow provides benchmarks for a few different models running locally. What did they find? Spoiler alert:
Impressed by these benchmarks, we’ve decided to upgrade our developers still using M1-based laptops to the new M4 Max MacBook Pros.
FWIW, I made the same decision, too (see the MKBHD entry below).
-
Why Hollywood Executives Are So Giddy About Trump 2.0: I still care deeply about entertainment and sports, perhaps unsurprisingly given how long I’ve worked in the industry. The money quote in this reporting comes from Warner Bros. Discovery C.E.O. Zaslav during an earnings call: “We have an upcoming new administration. It’s too early to tell, but it may offer a pace of change and an opportunity for consolidation that may be quite different, that would provide a real positive and accelerated impact on this industry that’s needed.” Whether that is better for quality entertainment vs quality earnings… I’ll leave that up to you.
Code & Tools #
-
First impressions of the new Amazon Nova LLMs: Amazon recently released a whole new set of models at re:Invent a few weeks ago. Simon Willison has a great rundown of the models, pricing and capabilities. He also updated his LLM tool to support the new models. That tool is really nice, btw - simple and has some nice features.
-
Skip: Skip allows Swift and SwiftUI developers to deliver apps for iOS and Android from a single codebase. Developers can add custom native Swift & Kotlin code if you need something unsupported by Skip. This will be great for the core UI of many apps, allowing Kotlin developers to focus on differentiated features. I’ve been keeping an eye on this since being introduced to the founders a year or so ago. Both founders are experienced devs on high profile apps with great reputations. The approach seems to make technical sense, as well. I just grabbed the latest SDK to start testing it out with an app idea I’ve been working on, am hopeful I can get an Android app running at the same time.
-
Backblaze Open Sources Boardwalk Workflow Engine for Ansible: I’ve started automating the setup of all of my servers in my homelab, learning Ansible along the way. This tool by Backblaze looks really interesting for horizontally scaled environments. The ESPN Fantasy Games services ran in a similar setup once-upon-a-time in small, sharded pods or clusters, which is why this tool jumped out at me. I could relate to the workflow they were trying to automate.
-
Scour: Neat little news aggregator built on vector search, RSS, and some cleverness. I came across it from the developer’s blog about efficient binary quantized vector embeddings, which look very cool indeed:
Vector embeddings by themselves are pretty neat. Binary quantized vector embeddings are extra impressive. In short, they can retain 95+% retrieval accuracy with 32x compression and ~25x retrieval speedup. Let’s get into how this works and why it’s so crazy.
Many years ago, a friend and I built a news aggregator that helped connect stories together as they evolved over weeks and months. The biggest side effect of all of this LLM excitement is a huge explosion in NLP tools. It’s easier than it ever has been to understand the meaning of text, even without using an LLM per se.
-
Tool Pre-Selection Using Embeddings: This POC and paper looked really interesting. There are vague similarities to the framework we built at Ohai, so if you’re curious how to think about architectures for more complex, sort-of open-ended assistants, this might be a good place to start.
Watch #
-
M4 Max MacBook Pro: I’m Convinced!: MKBHD’s review of the new M4 Max MacBooks basically convinced me to upgrade my daily driver laptop from an M1 Max to an M4, especially before any idiotic tariffs may be put in place. I’m writing this post on my new laptop with maxed out RAM, which was my only regret with the M1 Max it replaced. It’s wonderful.
-
Command: a new tool for building multi-agent architectures in LangGraph: While not exactly a new framework, this video was just released a week ago. It’s a good overview of LangChain’s take on multi-agent orchestration. There are a few of these multi-agent frameworks, including an exploration from OpenAI called Swarm.
-
Secrets of the NBA-TNT Divorce: Bill Simmons and Matthew Belloni talk about the NBA deal. Episode is 3 weeks old, so a bit “old news” but it’s a good episode covering the state of sports rights generally, and an interesting conversation regardless.