TJ

Closing the feedback loop for AI agents.
Left an industry after 5 yrs to focus on AI and impact

[TJ @agiprep]

nvm, a `closing 1 agent` being leftover and running eternally still happens about 10% of long-running tasks. But if this one gets addressed I think Codex really gets there now

[TJ @agiprep]

This has been completely fixed now,
and honestly it's now so `feature complete` in a way now it's getting harder and harder to find a fault. Amazing team.

[TJ @agiprep]

https://t.co/cmgPTPSJCQ

I cant add timestamps because it's still live but from the time of posting it's around 05hr 54m ago.

[TJ @agiprep]

I've never been this optimistic on South Korea in my life.

About 5 hours ago I watched a minister during a live cabinet meeting on TV that AI is not ready for adoption. His point was they had users use it and the results were poor. The PRESIDENT responded that an LLM's response quality varies according to how you prompt.

Then went on to explain that if you provide all the variables it responds well. Afterwards he goes on with such a high level of understanding of LLMs, I was speechless. Like he ACTUALLY understands the space in an application level.

Beyond politics, if you work with LLMs you know that this is such a profound level of understanding. FROM A SITTING PRESIDENT.

[TJ @agiprep]

This has been completely fixed now,
and honestly it's now so `feature complete` in a way now it's getting harder and harder to find a fault. Amazing team.

[TJ @agiprep]

Codex app is great until you have a 12+hr session running, and it's laggy af to switch to and from that convo

[TJ @agiprep]

I really need a search bar for Codex app settings. It's so hard to find what I'm looking for @thsottiaux

[TJ @agiprep]

This post encapsulates how my skills have been evolving into
plan -> implement -> Review loop for 10~20x the time it took to implement.

Should be the default, how to create a general-purpose review loop is the trump card

[Mario Zechner @badlogicgames]

recommended reading.

https://t.co/2GgZb04PoE

[TJ @agiprep]

This

[DHH @dhh]

I've had more "I can't believe it's this good" moments with GPT5.5 than any other model since Opus 4.5. It's shockingly, scarily capable. Days and days of amazing progress. All steering, no handwriting. Yet utterly delightful to conduct its coding. So, so good.

[TJ @agiprep]

Never seen Codex assign subagents as UUIDs before - temporary bug I guess?

[TJ @agiprep]

Aaaand it's fixed, appreciate it guys 🫡

[Austin Wang @austinywang]

@agiprep @lawrencecchen @manaflowai Should be fixed in v0.64.9 now. Sorry about that!

[TJ @agiprep]

@austinywang @lawrencecchen @manaflowai Ooh fantastic, updating now 🙏

I missed the notifications and shortcuts so much I realised I cant go back to iterm anymore haha

[TJ @agiprep]

After recent upgrade @manaflowai (cmux) is completely unuseable, I have about 12 tabs with 3 actively running ~2 claude sessions and it stutters to a complete stop after 10~15 minutes. If I ignore it it freezes my maxxed out M3.
cc @austinywang @lawrencecchen

[TJ @agiprep]

@lawrencecchen @manaflowai @austinywang I'll send you a DM

[TJ @agiprep]

@lawrencecchen @manaflowai @austinywang nvm I assumed the gh link was an unreleased version for me to try, yes my version is v0.64.8

[TJ @agiprep]

@lawrencecchen @manaflowai @austinywang hi @lawrencecchen yeah the symptoms did 100% feel like a memory leak somewhere. I'm on the latest version I will try this and let you know, appreciate the swift response

[TJ @agiprep]

Pretty consistently if I give @claudeai a task and it compacts the very next turn, it hallucinates what the task was.

This NEVER happens atm with Codex. It's a huge difference whether I can leave the agent to go wild with a long running task or not.

[TJ @agiprep]

https://t.co/KsqmbeTz71

[TJ @agiprep]

After multiple iterations including a rewrite into Python, the automator was added as an experimental module to the @BMadCode official repo. Install it to your repo via

`npx bmad-method@next install`

and be sure to select the automator during installation 👍

[TJ @agiprep]

Open-sourcing story-automator-go for the bmad implementation loop.

[TJ @agiprep]

@divya_venn The rest makes sense to me but why candlelight?

[TJ @agiprep]

No idea what happened to Codex after he edited those 3 files but looks like it switched to caveman mode?

Weird since I've never prompted or have any instructions that might trigger this 🤔

[TJ @agiprep]

Once you try to make your own voice recorder that also records speaker sound, you start to realise how magical google meet noise cancellation / voice isolation is.

It's not easy to get right even if you're only targeting your own machine.