Eng Coffee Sips EP.007
Rusty Rails
Eng Coffee Sips is a short factual based newsletter for articles that might interest engineers or engineering leaders, not just AI news. Short and to the point to skim over your morning coffee.
In this edition
Today’s edition is 838 words, a 4-minute read.
-
DHH tells Rails World it is pencils down on hand-written code, and HEY’s backend is agent-written Rust he has never read DHH opened Rails World by saying 37signals stopped writing code by hand a couple of weeks earlier. HEY is being rebuilt on a Rust mail server written by agents; he says it needs 99% less CPU and 95% less memory than the backend it replaces, knows no Rust, and evaluates the code as a black box.
My Take
This post isn’t designed to help you decide if DHH is good or bad. Do your own research before watching this video. Are these good points from a terrible person? That’s up to you to decide. Is DHH the modern day Steve Ballmer? What the hell does it mean to completely shit on a language on stage that you are moving your entire stack to? I have no idea.
( Rails World 2026 opening keynote / @dhh: pencils down / @dhh: ttfx ported from Rust to assembler / @dhh: the YouTube comments / @bendiken / @jpschroeder: the comments, collaged / @evilsocket / @adamhjk: learn to build with agents / Jared Norman: What About Rails? / Hacker News )
-
Anthropic and OpenAI ship new models, and both cut prices Anthropic released Claude Opus 5.5 at $4 in and $20 out per million tokens, 20% below Opus 5, and made it the Claude Code default; it says typical workloads run 40% cheaper at medium effort. The same day OpenAI shipped GPT-6 Sol and Luna, the tiers under Astra, at $2 and $10 for Sol and $0.10 and $0.50 for Luna, which it calls 50% below the GPT-5.6 rates and says is permanent. Artificial Analysis puts Opus 5.5 at $5.98 per task at max effort against $1.06 for Sol.
My Take
The harness wars continue! The competition here has been positive. We’re getting cheaper and better models quickly. I am starting to get model fatigue and I am sure others are as well. It is only the significant improvements I care about now and this feels like one so far. Basically I get Fable 5 at Opus prices? Yes please. Astra now similar to Fable as well in capabilities at cheaper prices? Great!
( Anthropic / OpenAI / Artificial Analysis / VentureBeat / @ParkerRex / @thsottiaux )
-
Minecraft’s creator says he was wrong about AI coding Markus Persson, who wrote Minecraft, posted in January that using AI to write code was ‘an incredibly bad idea still, and anyone advocating for it is either incompetent or evil.’ On July 21 he said he would try it, citing trouble hiring programmers; an August screenshot showed a Claude Code session scaffolding a project.
My Take
It is really interesting seeing these story arcs on AI. There’s a cliff that happens with a lot of people that I have noticed. At first things seem impossible to believe, then once you start to use it yourself, there is this holy shit AHA moment where you can really see what is possible.
-
Cloudflare has an unannounced Kafka-style event stream, K2, in its dashboard Screenshots posted by James Ross, CTO of game-host Nodecraft and a Cloudflare MVP, show a dashboard section called K2 Streams: named streams of event data with retention from one hour to 30 days, fed from a Worker binding or an authenticated HTTPS endpoint. Cloudflare’s public API schema now lists a k2 Worker binding, “available only to accounts enabled for K2”.
My Take
Cloudflare in the event streaming/kafka space? Very interesting as I build more and more on Cloudflare. It’s not quite a full featured platform but it’s getting closer and closer.
( cloudflare-typescript v7.2.0 / Cloudflare API reference / @CherryJimbo )
-
AWS releases Strands harness, an open-source agent harness it says costs 28% less AWS’s Strands team released Strands harness a general-purpose agent harness in one line of Python or TypeScript, on Bedrock, Anthropic, OpenAI, Google or Ollama models. The Strands CLI, built on it, shipped the same night. Its own six-benchmark run claims 28% lower cost than other harnesses at similar accuracy.
My Take
I dunno, this seems interesting but it feels like getting into the harness game or into the harness wars isn’t really that good of an idea. The harnesses change so rapidly that the notion of a better one that lasts longer than 6 months doesn’t seem realistic to me. Competition is good though, so perhaps its projects like these that keep this big vendors on their heels.
( Strands Agents / The Register / Hacker News )
-
Meta Connect puts the Muse agent on glasses and adds a keychain device for it Meta’s Connect keynote centered on Muse, the personal agent it launched two weeks earlier: voice, acting on what its glasses see, hands-free use, and Muse Charm, a keychain device for talking to it that Zuckerberg says ships in December, unpriced. Meta VR Glasses weigh about 100g at $1,299.99 for spring 2027, and a $149.99 FDA-cleared feature turns Meta Glasses into a hearing aid.
My Take
Personal agents keep popping up and Meta drops the updated Tamagotchi!!!!! I’ve tried OpenClaw, Instinct, and a private one. There is truly something to these but I’ve struggled to allow these to have access to much or any of my personal information at all. I keep seeing scare stories about things that have happened. I can’t see Meta being a good steward for a personal agent, but if my data was in there already I would be seriously considering this one. It looks pretty good. If you are interested in trying out Instinct, ping me for an invite.