Isaac Cochran

Software Dev | Codex Ambassador | Homelab Enthusiast

[Isaac Cochran @IsaacCochran12]

@NapVeg @thsottiaux It's linked at the top

[Isaac Cochran @IsaacCochran12]

@DevAdventur3s Doesn't do speaker recognition yet. But yeah, crazy

[Isaac Cochran @IsaacCochran12]

@Jchammond_ @jamonholmgren Yeah, they updated it to fix this after some people made a stink about it. Claude included the effort level in the initial system prompt for the model so when it changes it would invalidate the entire cache

[Isaac Cochran @IsaacCochran12]

@thsottiaux Anyone else notice the 2.1? That means there might be more to come!

[Isaac Cochran @IsaacCochran12]

@keleftheriou Excited to see it! If you need any help with testing, I've got a Pixel Watch 4 and have built apps for it in the past.

[Isaac Cochran @IsaacCochran12]

@DODOREACH @avedmala_ It's happened to me too, not something this big but let me know when packages arrive. I only gave it access to my gmail

[Isaac Cochran @IsaacCochran12]

@AbdullahAymanM @maria_rcks Agreed. We had the continue after usage abused, and now this??? Please no

[Isaac Cochran @IsaacCochran12]

Simply put, it's replacing creating individual threads yourself with one super thread that babysits and creates most of your other threads for you.

We're finally to the point where we can have one main thread that goes on for days and has a lot more context on what you're building (I. E. Dots, Grok Bot).

This thread then spins up your traditional coding agent threads to go and execute tasks on it's behalf.

Instead of managing the sidebar, I now talk to my dot, paste it a ticket link or describe a task, and then it makes sure it gets done.

Here's a rough rundown of my default flow:
1. Research: Codex + GPT 6.1 Sol
2. Code: Claude Code + Opus 5.5 (Called via CLI by Codex Agent)
3. Review: Codex + GPT 6.1 Sol
4. If there's any findings it repeats steps 2 and 3 until it looks good.

It's less stressful because you're not babysitting 7-8 agents, but instead you only need to respond to one.
And that agent will then supervise and ensure the other agents do their work properly

[Isaac Cochran @IsaacCochran12]

@JamesZmSun Would love to be able to spin up a cloud machine on my own homelab and give it to my bot.

The cloud machines are fine, but not great for the dev work I do (Need access to a bunch of tooling and mobile devices to test with).

[Isaac Cochran @IsaacCochran12]

@Wirenut3388 @thsottiaux Facts! I've made a temporary solution but would love official support

[Isaac Cochran @IsaacCochran12]

@Florgkbo @thsottiaux My dot has proactively messaged me about random things I didn't even ask it to. Kinda crazy

[Isaac Cochran @IsaacCochran12]

@yesadok @jpshrodinger The usage hasn't dropped yet

[Isaac Cochran @IsaacCochran12]

@JamesMalsawm @buildwithrajath Works on the $100 a month plan too, and I've been loving it with GPT 6.1 Sol on my $200 sub

[Isaac Cochran @IsaacCochran12]

@0xLalice @iannuttall Been an improvement for me. The benchmarks show the same. This is from the artificial analysis intelligence index:
https://t.co/PppIx6JvB5

[Isaac Cochran @IsaacCochran12]

@theo Really good benchmark. Good Job!

[Isaac Cochran @IsaacCochran12]

@StefanoGPT I think they silently brought it back a few weeks ago. I really hope people don't abuse it again this time

[Isaac Cochran @IsaacCochran12]

Dots are great for dev work. And when I gave it access to a bunch of stuff organizing my personal life too.|

Honestly it's not great at research tasks in my opinion but it can delegate other agents to do that

I still use ChatGPT for recommendations for things, but the bot is for getting real work done.

[Isaac Cochran @IsaacCochran12]

@atelicinvest Just use dots. Then have them use all the other stuff for you...