StartupHakk
Region: US
Monday 20 July 2026 22:54:46 GMT
16916
193
84
9
Music
Download
Comments
KJV Cards :
I spent $1k, mostly on Claude, this past month. It definitely isn't working like it used to. Codex can certainly be cheaper, but your plan has to have a high level of granular fallback, because the adaptive thinking in codex is poor, and GPT takes significantly longer to execute work. Retroactive token churn, within chat context, is the main issue, so you gotta open new threads, frequently. Sonnet 5 is the best ROI, right now, with light work on Codex, but in the next year, I'll be chucking then in the garbage for open models, fine tuning, etc, with actual workflow intelligence. 👍
2026-07-23 15:54:10
0
🌶️ustin :
Dude what are these people building!?
2026-07-21 04:21:12
26
pyFlummox Gaming :
my work account is the 200 dollar account and I literally have never been over 6% in a heavy session. these guys must be trying to rebuild the entire Internet 🤣
2026-07-21 04:59:28
30
Time To Wake up :
It's comedy watching these guys trying to make a broke business model work for them. $1000 to write code per month with no ROI.
2026-07-21 04:21:41
6
Fuck Google :
Why are you running 4 agent loops in parallel?
2026-07-21 04:39:34
4
idiot_finder :
Free tier ChatGPT for all of my free thinking and stupid questions, claude for precise coding and execution tasks
2026-07-21 13:45:05
2
NoteKnight :
Serious coding
2026-07-21 18:57:43
3
Departed Kitty :
Turns out it costs more than it’s worth. Lol.
2026-07-21 01:22:32
28
Merlin :
People burning through tokens like this, in general, don’t know what they’re doing, I think?
2026-07-21 19:58:34
1
𝕵𝖔𝖘𝖍 :
things haven’t been the same since fable went down after those first few days ngl
2026-07-21 04:31:01
0
wwf :
Been like that since opus 4.6 came out. Claude is unusable
2026-07-21 03:29:55
1
d_b123450 :
thats a prompt issue. if you use very specific prompts you can significantly reduce the level of reasoning and get exactly what you ask for.
2026-07-21 12:48:53
0
dressemups :
There’s a context bug. You can fix it in your terminal
2026-07-21 01:59:48
0
HydraTale :
you dont need to even be using agent loops, all people are doing is wasting tokens faster. just set better parameters
2026-07-21 05:00:34
4
Theatre :
Skill issue. I’m on the max x5 and rarely run into usage issues anymore.
2026-07-21 12:27:37
2
user5739179673719 :
Such a bandwagon. Good engineering practices and planning most of the coding can actually be done by Sonnet
2026-07-21 03:42:56
0
imtiredgoingtobedsoon :
agree with this take
2026-07-21 14:31:48
2
HeyJakeDotCom :
I’ve been trying to tell people this for months and they think I’m crazy lol
2026-07-21 21:13:27
1
michael.c.e :
4 agents for coding??? What he has a dedicated agent for adding comments to his code??
2026-07-21 12:52:47
2
Matt :
chatgpt wrote this
2026-07-21 14:00:24
2
100 :
yup! been like that for past 4 weeks!
2026-07-21 03:25:53
1
mikearoundthehouse :
No matter what I keep hearing, Fable keeps fixing the problems 5.6 Sol can’t. The tightest solution I found was Fable coordinator, 5.6 as the coders, opus for the tests.
2026-07-21 03:54:01
1
somaforthepeople :
Yep, I can barely use Claude now. Once everyone realizes Ai is the new computer it’ll probably get worse, then better again.
2026-07-21 02:22:45
1
aidan :
The price racks up.
2026-07-22 00:41:17
0
To see more videos from user @startuphakk, please go to the Tikwm
homepage.
