Harsh Vikram Singh

[Harsh Vikram Singh @Harshvikram14]

@theo @cansar Still behind on Frontend design? So I still can’t cancel my claude subscription?

[Harsh Vikram Singh @Harshvikram14]

@mattpocockuk Provenance. Seam. Load bearing.

[Harsh Vikram Singh @Harshvikram14]

@peduarte Enzyme migration complete

[Harsh Vikram Singh @Harshvikram14]

@thsottiaux Where are you, my reset?

[Harsh Vikram Singh @Harshvikram14]

@petergyang @ChatGPT I was thinking the same thing today. It is actually amazing that it just keeps on going!

[Harsh Vikram Singh @Harshvikram14]

Oh no! found a great use for @ChatGPT Work. put it to work, and in 5 hours, I am almost out of usage limits (Pro 20x). Thought I'll finally use my resets. And all have expired!! 😢

[Harsh Vikram Singh @Harshvikram14]

@DavidKPiano I made a big mistake! Went Opus 5 first, last 6 days. Now I dont even know which side-quest of the feature I am currently on!

[Harsh Vikram Singh @Harshvikram14]

@pghqdev @mattpocockuk Now, I am literally doing “yes to all” on the grilling tickets. Because, as you said, brain exhausted. Will probably revert all of it tomorrow

[Harsh Vikram Singh @Harshvikram14]

@pghqdev @mattpocockuk I liked it for vague, not super specific features. It got all the decisions, and branches correct, leading to nice feature implementation. And the route to implementation was in one case better than how I would have done it. But that was with Fable. Maybe Opus is the culprit ?

[Harsh Vikram Singh @Harshvikram14]

Thanks Claude

[Harsh Vikram Singh @Harshvikram14]

What is this phrasing called that is immediately recognisable as AI now:
"Two things I am flagging, rather than fixing",
"Here is the kicker",
"Here is how everything comes together".
What is this?

[Harsh Vikram Singh @Harshvikram14]

@i_mika_el @mattpocockuk That might be helpful. Or an actual UI, that can show decisions, and branches from those decisions, expected number of questions of a grilling session. Are you working on something that helps make these grilling sessions easier somehow? @i_mika_el

[Harsh Vikram Singh @Harshvikram14]

Hi @mattpocockuk , /wayfinder is wonderful. But there is one problem. Grilling. we need to make it easier. Everything is too verbose. I end up asking it to simplify it's explanations/questions way too much. I know it is absolutely required to make. but man, does the process suck!

[Harsh Vikram Singh @Harshvikram14]

@i_mika_el @mattpocockuk Here are the main pain points :
1. Verbosity of each question, leading to decision exhaustion
2. Lost context of why this is being asked, which decision triggered it, and what it aims to resolve. I end up asking those a lot, when the 'way' gets longer.

[Harsh Vikram Singh @Harshvikram14]

@mattpocockuk I have a few ideas, will share with you if I find any of those make the way-finding more delightful

[Harsh Vikram Singh @Harshvikram14]

I tried a Gauntlet loop on UI revamp using GPT5.6-Sol. The critic approved all its designs in first pass, and stopped. I can report that it absolutely sucked. Now will try it again with Opus 5

[Matt Shumer @mattshumer_]

I’m officially calling this the Gauntlet Loop.

The agent (not you!!) breaks the goal into parts, gives each part a specialist builder and a ruthless blind critic sub-agent, with a mandate to only pass if the generated artifact is better than some real-world equivalent.

[Harsh Vikram Singh @Harshvikram14]

Every 2-3 months, because of posts here, I’ll start thinking I can just one-shot big features. I set the /goal before bed, am happy for 5 mins the next morning. Start testing it. And then the whole day is ruined 😞

[Harsh Vikram Singh @Harshvikram14]

time to go 20x now

[Harsh Vikram Singh @Harshvikram14]

Thanks @thsottiaux 🥲

[Tibo @thsottiaux]

We're fixing a codex bug today that was causing us to undercount tokens being served to some Pro and Plus accounts by a small amount. This impacted < 15% of accounts.

Not the kind of bug you want us to fix, but didn't want to do this silently and thought you should know.