yoshi-taka sre

🙅 cd tab_completion_only_works_if_you_remember_the_prefix
🙆 cd *remember*

Zed now shows the publish date of a library version in package.json via tooltip.

just published
"How to Reduce AWS Lambda Cold Start Times by Shrinking Deployment Package Size"
ask-yoshi-taka.medium.com/how-to-reduc...

Looks like you can write code on Medium by typing triple backticks (```).

Aurora r8g is now supported across all AZs in Tokyo!

Rather than using an LLM to directly build features, you’re often better off building tools with an LLM.
Once the tool is built, it gives you better reproducibility and helps improve quality over time.

Debugging tools are a great use case for this approach.

Here’s a screenshot of the code:

Hi @quinnypig.com,

We’re running into issues with Aurora r8g in ap-northeast-1 (Tokyo). Some AZs aren’t available, and to make matters worse, it’s inconsistent.

new post

Table-Driven Programming: An Underrated Superpower
ask-yoshi-taka.medium.com/table-driven...

Remove unused dependencies by yoshi-taka · Pull Request #33189 · zed-industries/zed I verified it after running cargo shear. https://crates.io/crates/cargo-shear Release Notes: N/A

I love zed and cargo shear (and knip + e18e).

github.com/zed-industri...

Knip + e18e = cleanup goldmine.
github.com/yamadashy/re...

Which one is legit?

c15t.com

Since the official approach didn’t send cleanly, I manually mapped the Vercel AI SDK’s experimental telemetry to Langfuse.

Adding AI agent logs to PRs is great for traceability,
but for now we’ll have to attach them as files due to the description size limit.

Hit a FRAME_SIZE_ERROR with Rust gRPC when sending to Honeycomb.io, but it worked fine using HTTP + protobuf under the same setup.

Question: Are language servers struggling to keep up with the speed of AI agents?

If you’re using the same account for both personal and organizational work, it’s no surprise you can’t separate GitHub Copilot Premium Requests.

It looks like AWS SDK for Rust doesn’t support setting TCP keepalive.

Using enableCompileCache could slightly improve performance for Node.js Lambdas as well.
#AWS

The slowest step is compiling the Rust datadog-api-client.
#datadog #rust

Running multiple AI agents at once lets you push your brain and soul to the absolute limit. (Just kidding)

Agent Panel: click-to-update fails to preserve agent panel history · Issue #28735 · zed-industries/zed Summary Updating Zed via the lower left 'Click to update' option fails to preserve Agent Panel recent chat history Description When using the 'Click to update' option (lower left) during a coding s...

This is a small but annoying issue.
github.com/zed-industri...

aws-smithy-checksums transistive dependency fails cargo lambda build --release · Issue #993 · awslabs/aws-lambda-rust-runtime I'm not sure if this repo can do anything to fix it, but just today the latest aws-smithy-checksums dependency started using crc-fast and the release script for aws lambda rust runtime fails with t...

github.com/awslabs/aws-...

Use 'crc-fast' crate for better CRC performance by onethumb · Pull Request #4111 · smithy-lang/smithy-rs Motivation and Context The crc-fast crate provides better performance (>100GiB/s for CRC32 and CRC32C) and a unified solution for calculating CRC32, CRC32C, and CRC64NVME, the three supported CR...

This PR caused build errors for Rust Lambda:
github.com/smithy-lang/...

I love how this book explains table-driven programming.

The Performance Mode option has been removed from Amplify Console.
Honestly, it’s a good change.

trick2025/13-yoshi-taka at main · tric/trick2025 Contribute to tric/trick2025 development by creating an account on GitHub.

Totally forgot to mention, but I placed in the “Trick 2025” mystery code competition at RubyKaigi.
It’s my third time now!
Even when writing weird programs, I always try to keep the code as readable as possible.

Hope you enjoy the early-2000s vibe!
github.com/tric/trick20...

About premium requests - GitHub Docs Learn about premium requests and how they impact your usage and billing.

Is the premium request multiplier for o4-mini-high 0.33? There’s not been any documentation on it yet.
docs.github.com/en/copilot/m...

I’m having tests generated through vibe coding, and the coverage is high.
But since I have no idea what’s actually being tested, it’s unclear whether the tests are meaningful at all.

Vibe coding feels like human experimentation.
I’m just following instructions without learning anything, like a test subject.
It really makes me feel like an order-taker.