Then you look at the actual utilization and every instance is sitting at 2% CPU, 5% memory, and 5% bandwidth. So sad.

K8s is cool and it has its place. But sometimes a plain ASG with a handful of big boxes is enough.

Feels like a lot of teams go distributed way too early and light performance on fire for architecture points and the privilege of managing more instances, more moving parts, and more overhead.

So either there’s a trick I’m missing, or the only supported setup blocks the tuning S3 Files needs to perform well.

In either case, it’s rather frustrating :/

But S3 Files doesn’t support ECS EC2 launch type. That leaves ECS Managed Instances, where you can’t tune kernel params. So the one fix it needs is blocked by the only setup it supports.

TIL: S3 Files needs sunrpc tuning for decent throughput, especially on a single machine. Without it, you won’t come close to the throughput most boxes can deliver.

ECS Managed Instances make me very very sad.

@zed.dev this is a lot to take in all at once. 😅 so many changes!

Refactor ordered content runtime to execute scan misses by ahrav · Pull Request #308 · ahrav/Gossip-rs This commit overhauls the ordered_content module in gossip-scanner-runtime to more accurately reflect its responsibilities. Previously, the module handled both page acquisition and direct local fil...

Surprisingly little lifetime or borrow checker noise. Most feedback is about real correctness or edge cases. Even when it’s wrong, I can usually see the reasoning. 6–8 months ago it felt random and frustrating. Big improvement imo.

Example: github.com/ahrav/Gossip...

This is all based on Rust-heavy use. Bugbot hasn’t worked well for me so far, but I haven’t figured out why yet.

Coderabbit feels more focused on the core review experience and has found a nice balance.

KiloCode is still very strong, but it’s also doing more across triage, security, and other workflows. That wider scope might be why the difference shows up more in the overall ranking than in code-only.

Code-only (just finding issues in code, ignoring extras):
KiloCode > Coderabbit > Codex > Greptile > Gemini Code Assist > Claude Code > Cursor Bugbot

Here the gap between #1 and #2 is much smaller. They feel closer in raw code review quality.

I’ve tried a bunch of code review agents the past few months, mostly Rust.

Overall (features, UX, extras):
@coderabbitai.bsky.social > @kilocode.ai > @greptile.bsky.social > Codex > Gemini Code Assist > Claude Code > Cursor Bugbot

There’s a clear gap between #1 and #2 here.

Code-only in thread

Screenshot of an AI message saying “oops that wasn’t a dry-run build”

What… no, it’s not fine! 🤦🏼‍♂️Dry-run exists for a reason. Welp, I guess we’ll see what breaks in 20 minutes…

Umm… what? I’m not sure what this means.

TIL there is a thing called causal profiling 🤯 Amaze-wow!

Betterleaks, a new open-source secrets scanner to replace Gitleaks A new open-source tool called Betterleaks can scan directories, files, and git repositories and identify valid secrets using default or customized rules.

www.bleepingcomputer.com/news/securit...

Oh this is fun. My LA brain is not fully comprehending what this means.

Hey! I’m in that blog.

New 'Zombie ZIP' technique lets malware slip past security tools A new technique dubbed "Zombie ZIP" helps conceal payloads in compressed files specially created to avoid detection from security solutions such as antivirus and endpoint detection and response (EDR) ...

bleepingcomputer.com/news/securit...

I’d rather write yaml than deal with CDK nonsense. 😖

Thanks @coderabbitai.bsky.social. Just the right touch of silliness to bring a smile to my face :)

"🐰 I hopped through slabs and frames today,

Tokens tucked away, gone astray —
Split-points whisper, quiet and neat,
No more pages, just ranges to meet.
The rabbit cheered: simpler beats complete! 🥕"

Codex is incredible at writing code, but soo bad at almost everything else.

It writes some code.

Me: “Err… I don’t get it. Can you explain?”

Codex: “Sure.”

Then it proceeds to explain the code… with more code.

Me: “No worries, Codex. I’ll ask Claude instead. Thanks for the hieroglyphics tho” 😂

Go slow, to go fast. I feel like I have to repeat this to myself multiple times a day.

GitHub - betterleaks/betterleaks: A Better Secrets Scanner built for configurability and speed A Better Secrets Scanner built for configurability and speed - betterleaks/betterleaks

github.com/betterleaks/...

that new new

SANDWORM_MODE: Shai-Hulud-Style npm Worm Hijacks CI Workflow... An emerging npm supply chain attack that infects repos, steals CI secrets, and targets developer AI toolchains for further compromise.

socket.dev/blog/sandwor...

Solid Friday afternoon!

Rare Not Random Using Token Efficiency for Secrets Scanning

lookingatcomputer.substack.com/p/rare-not-r...

“I’m honestly not sure how plan mode handles this. The concern is valid — if the planning phase and execution phase don’t share the same context, the executor might not know to update beads status.” - Claude (Opus 4.6)

“I’m honestly not sure…”. I’ve never been this happy to get a rejection. 😍

So many little rascals! ❤️#Caturday

Tiny dumpster card with words of motivation.

Thanks Tiny dumpster ❤️