Onkar Jadhav

Software, AI, and whatever catches my attention long enough to become a project.

[Onkar Jadhav @leo4dev]

@thsottiaux 1. Workflows from claude
2. Lower RAM use
3. Better sidebar pls. Why is it that the newer chats and projects are below and based on the day added and i have to drag them up.

[Onkar Jadhav @leo4dev]

@thdxr Earlier or even now the subagents spawn with all parent context right?
I not sure if they fixed it. But I mostly use codex through claude code. I love workflows.

[Onkar Jadhav @leo4dev]

@thsottiaux Codex with T3 code. Light and intelligence together.

[Onkar Jadhav @leo4dev]

@sickdotdev Exist

[Onkar Jadhav @leo4dev]

@lightsilver323 1. Stride: AI-based wellness app
https://t.co/UU2obyzkGE
2. Vaani: WisprFlow Alternative
https://t.co/rQMMPUOdVM

[Onkar Jadhav @leo4dev]

@DavidOndrej1 It's not Pro, this plan sits between Free and Pro.

[Onkar Jadhav @leo4dev]

@thsottiaux Saint Tibo, did you forget about banked resets?

[Onkar Jadhav @leo4dev]

@mattpocockuk Finally.
Did you notice any lack of depth in questions through batches?
In the earlier version of the skill it said some about multiple questions being bewildering.

[Onkar Jadhav @leo4dev]

It was my reset day😭

[ClaudeDevs @ClaudeDevs]

We've reset 5-hour and weekly rate limits for all users.

[Onkar Jadhav @leo4dev]

@theo You should a video showcasing new additions to T3 code. It gets confusing with so fast development.

[Onkar Jadhav @leo4dev]

@mattpocockuk Does this get the same depth it got before?

[Onkar Jadhav @leo4dev]

@thsottiaux GOAT!!!

[Onkar Jadhav @leo4dev]

I recently tested GPT-5.6 Sol in high reasoning mode, and the first thing I noticed was the speed. It feels much faster than the previous generation.
The UI and frontend capabilities have improved a lot too. The taste improvements are real and immediately visible, although it is still nowhere near Claude on that front.
I have not tested fast mode myself yet, but the reports from other users look promising. We also have not seen the 750 TPS Cerebras run yet, so I expect it to get much faster.
The reasoning is noticeably better. That said, Sol still does not reach Fable's level of understanding user intent, especially when it comes to picking up on what the user did not say. Fable is excellent at that. Sol still misses some of it.
The coding has definitely improved. It comes up with better feature ideas, handles long tasks well, and can keep going through very, very long sessions without falling apart.
I even used Codex to create the attached video. I will keep using the Codex app, and the result of this test is below.

[Onkar Jadhav @leo4dev]

GPT-5.6 Sol is good at naming stuff. It just spat out great names for app ideas i had.

[Onkar Jadhav @leo4dev]

@shsjshilyish @kirhgoff @mattpocockuk Sometimes. If it's small bug then no, I just tell it to fix it. If the agent has misunderstood the functioning of a feature then I grill again.

[piq and 69 others @piq9117]

@github they have to ship you CDs because the website is barely up

[Onkar Jadhav @leo4dev]

As you add features I think you should grill. And save them in docs.
Grilling is for planning out a feature or product. It helps you plan in detail and agents are great at asking questions that you wouldn't have even thought about. If you feel after few sessions that the agent forgets you goal, just re grill and save it to docs or AGENTS.md
My flow is to grill, I ask the agent to ask 5 questions per round. After shared understanding is reached, the next prompt is /to-prd which then followed by /to-issues. Both stored locally.
After all are implemented you can delete them. This ensure that features are implemented as I wanted them to be.

[Onkar Jadhav @leo4dev]

@ClaudeDevs FUCK.
My claude weeks started 6 hours ago.
Basically this is a waste reset for me.
We should have an feature to bank resets like codex.

[Onkar Jadhav @leo4dev]

@thdxr How's your experience?
How do you find the reasoning and token usage?
Any standout moments so far?