hyukki

[hyukki @hyukki725]

@_swanson @fractaledmind Great listen! One of the best engineering podcast episodes in recent memory.

[hyukki @hyukki725]

@tushargugnani @wesbos Chess puzzles is such a great idea. Are you using a service that already has this or building it yourself?

[hyukki @hyukki725]

@trq212 How did you get it to create a trailer?

[hyukki @hyukki725]

@rails @flavorjones What was the original talk supposed to be about?

[hyukki @hyukki725]

@mattshumer_ These agents have auto memory features which sort of overlaps with this. Do you think this is a lot better than using auto memory?

[hyukki @hyukki725]

@thsottiaux Is there a way to revert code per prompt? The /btw feature in Claude code is interesting and I’d like to use it in Codex CLI.

[hyukki @hyukki725]

I tend to get better results with Codex, especially for code review or finding bugs.
When you were comparing the two, did you have the same rules/skills setup for both? Also, Claude has auto memory and Codex has a similar feature in beta that has to be enabled which can also account for some differences.

[hyukki @hyukki725]

@LLMJunky This is incredible. Great list of use case ideas.

[hyukki @hyukki725]

Codex CLI has been my main driver for a long time now, but it's been getting so much better. Great work! @thsottiaux @OpenAI

1. Voice transcription (hold space to talk)
2. Best bug finding code review on the market (/review)
3. Automatic parallel subagents
3. Auto user memories with diff-based forgetting.

The diff-based forgetting logic is fascinating.

[hyukki @hyukki725]

@danshipper @every @kieranklaassen @kieranklaassen I'd like to try using the plugin. Would it work to only use the compound step of the workflow at the end of a feature? I get great results using Codex CLI vanilla with minimal rules and skills and wanted to keep my setup fairly minimal.

[hyukki @hyukki725]

@_swanson Why auto? It seems like most devs these days are using Opus 4.5 or 5.2 Codex xhigh.

I'm using Codex CLI + GPT 5.2 Codex xhigh and it's been great. Slow but accurate.

[hyukki @hyukki725]

@yongfook I agree it works pretty well vanilla, but you’ll save time and money when you’re constantly having to repeat your conventions and preferences.

[hyukki @hyukki725]

@bcherny What about skills? Do you add and use any custom skills?

[hyukki @hyukki725]

@_swanson @fractaledmind Amazing! @daisyui_ is this on your radar? Using Invoker Commands to open/close modals without any JS?

[hyukki @hyukki725]

@jorgemanru Why did you decide not to go with sqlite?

[hyukki @hyukki725]

@jorgemanru I've learned more from the code walkthrough that @dhh did on Campfire than any programming book or course and I was hoping there would be something similar for Fizzy. Looking forward to it!

[hyukki @hyukki725]

@gdb Is it possible to review in demand instead of automatically reviewing every PR?

[daisyUI @daisyui_]

daisyUI has ZERO dependencies by the way.

[hyukki @hyukki725]

@flavorjones great talk on ActiveRecord Tenanted!

Do you have any recommendations on how to migrate to ActiveRecord Tenanted from an app that's using acts as tenant?

Also, what's the best way to query across tenants for analytics purposes?

[hyukki @hyukki725]

@ericzakariasson @cursor_ai I try to use apply intelligently for most use cases by default to minimize context bloat