Skandesh

building boring profitable apps @lowkeytek

[Skandesh @skandesh]

claude code wrapped

thanks @AnthropicAI

[Skandesh @skandesh]

@thsottiaux why tho??

[Skandesh @skandesh]

claude models have become dog-shit after the whole american govt intervening into the safety checks.

its too consicous on following the guardrails rather than the actual instruction.

gpt>>claude

[Skandesh @skandesh]

@NoahShreve DM

[Dhravya Shah @DhravyaShah]

I do not fear failure. I fear mediocrity.

I will either end up building billion dollar company(s) or be homeless

[Skandesh @skandesh]

@thsottiaux honestly love the way it locks into the task and strive hard until its met.

5.6-sol-xhigh is goat'd but the harness(codex/chatgpt app) is trash, very memory hungry :(

[Skandesh @skandesh]

gpt 5.6 sol xhigh fast is absurd, burnt through the weekly limit in an hour.

[Skandesh @skandesh]

claude code workflows is way better clean implementation of orchestrating sub agents

[Skandesh @skandesh]

codex harness is the worst to use gpt 5.6 sol for.

last 2 days the performance of codex app, cli has been exponentially degraded.

it was consuming ~50-60GB of memory to just run one session with gpt 5.6 sol xhigh (not ultra)

and the codex process with 5.6 ultra spawns sub-agents like running fork within while loop which are unecessary while not asked by the user and the worst part is it doesn't get cleaned up..

[Skandesh @skandesh]

@aibuilderclub_ it doesn't work reliably, got 429 too often

https://t.co/s0gyC9r8VN

[Skandesh @skandesh]

bro wutt

[Skandesh @skandesh]

bro wutt

[Tibo @thsottiaux]

If you aren't yet bold enough to install the Codex app, you can stay in the presence of your orange crab and point it at GPT 5.6 Sol. Takes 5 minutes. Kudos to Theo for explaining one of the ways to get this done.

Step 1: Install CLIProxyAPI
Step 2: Connect
Step 3: Define following alias and enjoy claudex

```
alias claudex='CLAUDE_CODE_SUBAGENT_MODEL=gpt-5.6-sol \
CLAUDE_CODE_ALWAYS_ENABLE_EFFORT=1 \
CLAUDE_CODE_MAX_TOOL_USE_CONCURRENCY=3 \
ENABLE_TOOL_SEARCH=false \
claude --model gpt-5.6-sol'
```

If this gets blocked, I owe you a reset.

[Skandesh @skandesh]

cursor vs grok cli lol

[Skandesh @skandesh]

@VittoStack love to join

[Skandesh @skandesh]

@pejmanjohn @trevin @kieranklaassen ++ to this feature

[Skandesh @skandesh]

claude code's full screen renderer is horrible.

- unexpected clicks turning into clicks (accept a decision while reading through the plan)
- laggy while scrolling huge sessions
- consumed more CPU than usual session

[Skandesh @skandesh]

@iuditg @OpenAIDevs @OpenAI @thsottiaux been feeling this 3 days before posting this and still now

https://t.co/YVdXrfrZHD

[Skandesh @skandesh]

@thsottiaux i've felt after the manual reset the limits are draining so fast..

did a manual reset yesterday and already consumed 35% of my weekly usage.. strange

[Skandesh @skandesh]

opus 4.8 feels soo dumb rn

- keeps on forgetting mid session
- doesn't follow the instruction
- keeps inventing stuff without actually reading the code ( too lazy)

@claudeai bring back fable!!

[Skandesh @skandesh]

@ben_burtenshaw looking forward to this

[Skandesh @skandesh]

@thsottiaux i've felt after the manual reset the limits are draining so fast..

did a manual reset yesterday and already consumed 35% of my weekly usage.. strange

[Skandesh @skandesh]

https://t.co/09lzTTDz3I

[Skandesh @skandesh]

built a tracer viz!!

i've been tokenmaxxxing a lot recently with codex and claude and sometimes i wanted to see in-depth analytics of all the sessions and tokens spent per request.

wanted to see which harness performs better, but coudln't find a decent tracer viewer..

[Skandesh @skandesh]

seems like this could be a viable product if positioned well..

[shadcn @shadcn]

anyone built a reliable system/solution for indexing and searching threads across claude code, codex, cursor..etc? All the built-in ones are not that great.