HNHacker News
TopNewBestAskShowJobs

azuanrb

398 karma · joined October 4, 2022

I'm a software engineer based in London, originally from Malaysia (we have great food!)

https://azuanz.com

submissionscomments
azuanrb··on Shopify is moving from React Native back to Swift and Kotlin
I’ve worked with both native and cross-platform. I think the mentality of being able to make changes quickly without much review often comes from cross-platform, especially when the developers come from a web background.

Changes are cheap and fast, so teams often feel less pressure to test everything thoroughly before a release. Which is a fair tradeoff. That’s part of the reason we can have dozens of releases a day on the web. Not just because we can, but because sometimes we have to.

With native, you know each release is harder to roll back, so you tend to build more tooling around releases, think through changes more carefully, and test more thoroughly before they’re ready to ship. You opt for one bigger, more stable release every few weeks instead.

At the end of the day, both approaches work.

Heh, now that I think about it, maybe web and cross-platform devs were the original vibe coders? Changes are cheap and fast. Just move fast and break stuff.

Native devs are the old-school ones. Shipping is expensive, so you better get it as right as possible.

azuanrb··on Google Antigravity TOS: 3rd party usage can get Google account suspended
It’s detectable and has been known to result in account bans. People have been doing it since the early days of Gemini, when the ToS was still unclear, and some only found out after their accounts were banned. There have been multiple GitHub issues and reports to Google about it. It’s not worth the effort unless you’re willing to risk a secondary Google account.
azuanrb··on Ask HN: Advice on Migrating from 1Password?
This anti-DHH movement is so weird to me. People seem to read summaries of his views, often based on interpretations that get exaggerated from one article to another until they become something else entirely. At least read the articles yourself, including his other work, so you don’t just cherry-pick the ideas that conform to your existing beliefs. Then form your own opinions.
azuanrb··on Superlogical Pre-Alpha Demo [video]
With the recent videos, I think there are multiple selling points.

* General performance. * Run the server and client everywhere. There are already multiple libghostty clients for mobile. Native terminal support on mobile is great, especially for coding agents. * Native sessions. The lack of sessions in terminals is part of the reason why some people went with tmux in the first place. * Multiplexer performance isn’t noticeable until you start running multiple sessions with multiple agents. Sometimes half of my RAM ends up being allocated to Ghostty/Alacritty + tmux. * As for rendering, performance is one thing. But the multiplexer tax is really, really annoying. For example, some features are available in Ghostty but not tmux, and vice versa. It’s the little things that pile up.

Another point is that, for some reason, multiplexers and terminals often diverge in terms of feature implementation. Probably because one traditionally focuses more on remote workflows, while the other focuses on local workflows. With AI, though, it makes a lot more sense to have both. Running long-lived sessions with agents is becoming normal, so the line between those use cases is getting thinner. That’s probably why they’re focusing on this approach.

azuanrb··on Mamdani Bans AI in NYC Schools
Sure, if the kids want to learn. But let’s be honest, most of them, or even us, are just going to use the chatbot directly for the answer.
azuanrb··on Claude Fable 5.1 and Claude Mythos 5.1
We rarely upgrade our phones or MacBooks because the newer version can do something the previous one literally couldn’t. Often it’s the efficiency, speed, battery life, etc, combined, that lets us push the hardware further.

I get your point, but we can only have groundbreaking leaps once in a blue moon. That doesn’t mean incremental improvements aren’t useful.

azuanrb··on “It works better in the app”
Imo, you already lost the argument when you have to explain that PWA is the "app". For most users, app is something they install from App Store or Google Play. I get the point of view from developers point of view, but unless the majority of the users are technical people, there's no reason to use PWA these days.
azuanrb··on Small Models Have Arrived
I’m building an internal tool for our company, basically an agent to help with on-call and alerts via Slack. I have evals running across a few scenarios, and my favorite models so far are Sol medium and Luna xhigh.

Sol medium has been a nice balance between intelligence and response time. Luna xhigh can achieve similar scores on the evals, but it takes noticeably longer. My impression is that the higher reasoning effort helps compensate for the lower base intelligence.

Cost is definitely a big factor, but latency and intelligence matter too. If I had the budget, I’d take Sol medium over Luna xhigh.

From using both on real scenarios, Sol is noticeably better at navigating around issues, exploring alternatives, and being creative when the obvious approach doesn’t work. That matters quite a bit when you’re investigating live alerts, where the path to the root cause isn’t always straightforward.

azuanrb··on A week of using Codex more than Claude
I’ve been experimenting with this for a while, and right now I’m using Luna xhigh as my default. Previously I was using Sol medium.

Sol medium is a great balance between speed and being thorough, but it’s quite expensive. Luna xhigh seems to compensate for slightly lower intelligence by thinking and reasoning for longer, so tasks can take more time to complete. But it’s crazy cheap.

I also have some custom evals using promptfoo to make sure I’m not introducing regressions when switching models. So far, Luna xhigh has been really, really good for the price.

Don’t sleep on it. Give Luna a try.

azuanrb··on Kino: A high-performance Ractor web server for Ruby 4.0
People love taking generic engineering problems and somehow spinning them as Ruby specific issues. None of the issues you mentioned have anything to do with Ruby. They’re problems you can run into in any engineering project.
azuanrb··on Ask HN: GitHub employees what's going on? Why?
It’s 2026 and we’re still benchmarking Ruby using Twitter’s 2008 architecture. It was less about a language choice and more about years of architectural redesign. I thought we were past this by now.
azuanrb··on OpenChamber: An Agentic Development Environment
> running them on terminal

That is the value. You can run them remotely from your phone and set them up on a headless machine too. You’re not bound to your terminal.

My usual workflow is to still use the terminal for heavy work. But sometimes I just want to prompt from my phone and run a few tasks or do some planning. Then I can simply resume the session in the terminal later. It's the flexibility.

azuanrb··on OpenChamber: An Agentic Development Environment
It’s just a personal preference, to be honest, so I wouldn’t overengineer it yet. In my case, I run it in a VM behind Tailscale. The main reason is that I want it to be able to install or use whatever tools it needs without contaminating my main machine. It can expose anything I need over Tailscale. So for me, it’s basically just a cheap, throwaway machine.
azuanrb··on OpenChamber: An Agentic Development Environment
I prefer Paseo https://github.com/getpaseo/paseo, mainly because I have specific preferences for different harness + model combinations. For example, I like using ChatGPT models via pi, and GLM via Claude Code.

If you’re happy with OpenCode as the harness, OpenChamber is great. But if you prefer using different harnesses under the hood, Paseo is a better fit.

I installed it on my homelab and can access the same sessions from my MacBook or iPhone at any time. Been really nice to use so far.

azuanrb··on Shopify replaced Redis with MySQL for inventory reservations–and it scaled
Durable workflows are different, not necessarily simpler, imo. Unless the team is already familiar with them, I wouldn’t introduce one just for this. You also have to account for the infrastructure needed to run and manage the durable workflow itself, which adds complexity.
azuanrb··on Shopify replaced Redis with MySQL for inventory reservations–and it scaled
Pretty interesting read. One thing I’m curious about is the DB size trade off. Going from a quantity in Redis to one row per reservable unit seems like it could create a lot more rows, even with the 1,000 row cap per item/location.
azuanrb··on Timeline of the OpenAI accidental attack against Hugging Face
Both can be true. How often do we hear about hacks that ultimately came down to bad defaults or simple security mistakes? That doesn’t mean any script kiddie could have discovered and exploited them.

These things often look obvious and simple after the fact. Finding the weakness in the first place is the hard part, and that’s what makes the agent’s capabilities interesting here, especially at scale.

azuanrb··on Pi's Minimalism Is Its Advantage
Never learned Emacs but am a Neovim user. But similar to what I said to my colleagues as well. Codex, Claude Code are VS Code, Jetbrain. Pi is Neovim.
azuanrb··on Cursor removed cost information from the usage page and CSV export
Thanks! I ended up built my own too. Just thought there are other better options out there that I might've missed.
azuanrb··on Cursor removed cost information from the usage page and CSV export
What tools are you using to run the comparison? Or are you just running the prompts and checking them with something like ccusage?

I’m asking because I’ve been looking for agent harness comparison tools too. I’m interested in more than just the inputs and outputs—I also want the system prompts, traces, and tool calls. It’s useful to understand why Codex, for example, uses more tokens while Pi doesn’t.

Fewer tokens aren’t necessarily better if the agent skipped important checks. On the other hand, using more tokens could just mean it’s overthinking the process. Either way, seeing the full execution trace for the same task is really valuable.

azuanrb··on qm – Multiplayer agent harness for work
I'm currently using it to help me with my oncall, first responder to our any production alerts. It's not as efficient as coding agent by default, but it's been tremendously helpful to me.
azuanrb··on Superlogical
That explains why he built sessions into libghostty instead of Ghostty itself. Looking forward to it.

I spend most of my time in the terminal and have been using tmux to manage my sessions and workspaces, mainly via sesh.

But recently, I decided to move to Kitty and use its built-in sessions instead. I built something around it as well. It’s still early, but I’m really liking this workflow so far.

https://github.com/alienxp03/kesh

azuanrb··on OpenAI reduces Codex Model Context Size from 372k to 272k
Curious, what kinds of tasks do you do that require such a large context window? Anything specific?
azuanrb··on GPT-5.6
It is a valid concern but I've been running yolo mode since the inception, using Claude and now Codex. I'm not bragging or anything, since I'm feeding my own curiosity too, trying to answer what's the worst that could've happened? So far, nothing catastrophic that I recall.
azuanrb··on Immich 3.0
The 2 days were spent uploading my entire library from scratch. After that, daily syncs only take a few minutes.

I don’t think this should be framed as iOS vs Android. From a technical perspective, I understand why the behavior differs. But from a user’s perspective, all they see is that one app works much better than the other.

My family doesn’t care about the technical reasons. They just saw one app finish much sooner than the other, and they preferred the one that worked better. Based on that experience, I can’t recommend Immich to them.

azuanrb··on Immich 3.0
No, it’s an Immich issue. Not OP, but I was already using Synology Photos. Synology finished syncing my library in two days, while Immich was still syncing after more than a week. I decided it wasn’t worth it after that experience.
azuanrb··on Herdr: Agent multiplexer that lives in your terminal
I'm using tmux + zoxide+ https://github.com/joshmedeski/sesh. Then on Ghostty, I have keybinds for my workflow. Cmd+k to open/switch workspace. Each workspace is just a new tmux session.
azuanrb··on Domain expertise has always been the real moat
I recently reviewed an app built mostly with vibe coding. The owner said it was almost ready to launch and just needed a quick check.

After looking through it, the database design was a mess. Some features worked, some didn’t. I explained the missing pieces and why things were breaking. Like OP said, he’s the domain expert.

I used billions of tokens last month alone. The tools are getting better fast. But giving AI to a domain expert doesn’t mean you no longer need software engineers.

A domain expert can use AI to build software. And a software engineer can use AI to learn about the domain. Both bring different expertise to the table.

azuanrb··on Orchestrating AI code review at scale
I’ve built something similar internally, but under the hood it’s mostly codex exec + Git worktrees. The main advantage over diff-only review is that it can walk the entire codebase, trace dependencies, and understand architectural or cross-system impacts instead of only looking at the changed files. The tradeoff is that it’s noticeably more expensive to run. I'm still experimenting on it but I quite like this approach so far.
azuanrb··on The last six months in LLMs in five minutes
It depends on what you’re comparing it against. For $20, OpenAI is still probably the best value for SOTA models. In terms of limits, you can use GPT-5.4 instead of 5.5. The intelligence feels similar, but it’s cheaper. You can also experiment with other harnesses like pi. It’s lightweight but capable enough, and its token usage is definitely much more efficient.
← PreviousPage 2 of 5Next →