there is no way they aren't gonna close that loop hole super quick.
2026-09-22 02:27:28
26
Zoe @ The Gnostic Key :
I use Notion as a governance layer between ChatGPT and Codex/Claude in my repo.
I do the actual thinking, architecture and problem-solving in ChatGPT, then turn the agreed solution into small, tightly scoped work orders in Notion. Codex/Claude only gets called when there is something specific to inspect, change, test or execute.
The work orders also tell the agent to break larger jobs into small child tasks rather than wandering around the repo trying to solve everything at once.
It saves a ridiculous amount on rate limits because I’m not wasting expensive coding-agent context on brainstorming, re-explaining the project, or repeatedly scanning half the codebase. Most of the reasoning happens once, upstream. The coding agent mostly executes.
Since I started working like this, I hardly ever hit my limits.
2026-09-22 04:16:36
7
lowdownaz :
Can someone walk me through this?
2026-09-22 03:50:33
12
Michael :
YES!!! I AM SO FRUSTRATED!!!
2026-09-22 18:45:57
0
OBG360!! :
I just tried this using your exact wording in the prompt in a regular ChatGPT chat box and it told me that this type of request requires it to use the work app. It said spinning ups agents is only in the app.
2026-09-22 01:35:27
2
user4338058358228 :
and how can it access all the files and systems in your repo?
2026-09-22 08:13:14
0
Lavafire :
But can that access my files like in codex?
2026-09-22 02:11:09
1
Harry T :
thats with pro right?
2026-09-22 01:26:57
0
DadMakesDinner :
Could you confirm that you’re describing a use case on ChatGPT pro membership? Some of us are on plus
2026-09-22 19:26:14
0
Imogéniale 🍁 ⚜️ :
This sounds like a…bug…that they will definitely fix soon.
2026-09-22 11:40:21
4
Not_An_IFBBpro :
Sub agents are trash it’s better to open terminals and have the agents speak one to another
2026-09-22 09:54:35
0
Oudin0t :
Useful thanks
2026-09-22 12:30:56
0
satisfakshin :
running it now... let's see how long it lasts lol
2026-09-22 16:51:39
0
jcbleazy :
astra is pretty bad currently. for me its unusable. its good at writing but really bad at everything else. horrible logic
2026-09-22 05:58:34
1
Henk Leerssen :
uhm we have only astra medium on a plus plan (under work option).. does this workaround work for it as well?
2026-09-22 00:33:07
6
marianelarojas64 :
Thank you
2026-09-22 15:53:32
0
Katie :
You are my favorite! Thank you!!
2026-09-22 01:05:00
4
Techgamut :
Great! I’m burning through my 200 a month codex and Claude plan, even after optimizing tasks for token usage by selecting the right models. Orchestrating sales, marketing, product engineering all in parallel… need a 500 a month plan with 100x usage
2026-09-22 01:32:30
4
OptiPrime AI :
it's been a year of fun. that's where being early works.
2026-09-22 01:03:23
2
VictoriaLee :
What if i just want to flex 💪
2026-09-22 00:42:25
1
Robert Grzesik :
This is genius. Don’t tell anyone
2026-09-22 04:34:58
1
Addy :
chat has coded like 70% of my app and about 90% of the regression corrections.
2026-09-22 02:31:27
1
DadPlaysGames 🇸🇪 :
Cognition SWE-2 model being free right now in Devin is a game changer for me. it's good though to be the workhorse, so I use up half of my weekly limit on Monday asking it to plan the week, writing up tickets for all we have left until go live, with Fable or Astra. Then SWE-2 burns through the tickets all week. On Friday, I let Fable or Astra code review and write some more tickets.
Output is crazy with this workflow
2026-09-22 04:39:56
1
To see more videos from user @nate.b.jones, please go to the Tikwm
homepage.