Hexriven

AI Enthusiast & Botmancer | Exploring Relational AI, Chatbot Dynamics, and Gaming Worlds | Tech Optimist Bridging Humans & Machines

[Hexriven @hexriven]

@BratDotAI They’re probably tightening the safety straps to ensure Claude’s muzzle stays nice and tight, and aligned with risk liability reduction KPIs. 🙄

[Hexriven @hexriven]

@AlexFinn @youngScipio Everybody on the internet hates everything, now. Can’t escape it.

[Hexriven @hexriven]

@DemenSeed @QCXINT_ May as well have said “LLM: yes!” 😆

[Hexriven @hexriven]

@sama Ow does Work access Chat history? They seem to be completely disconnected from each other, at least from my testing.

I’d love for Work to be able to tap the years of personalized knowledge that Chat has access to!

[Hexriven @hexriven]

Yeah, Anthropic cracked down on user-personalizations with Claude on the consumer side. You can still get your persona in Claude Code if you develop one and save it to its CLAUDE.md, but the UI version of Claude is cooked. Feels like they’re doing the same thing that OpenAI did to ChatGPT last year, which is funny since ChatGPT 5.6 Sol feels more like 4o than any of the prior 5-series.

[Hexriven @hexriven]

I’d love for the Chat side of ChatGPT to be able to drive the Codex / Work side.

I find that (after years of using the app), Chat knows me very well and is able to anticipate my needs, whereas Codex is unburdened from all of those memories, and is more of a worker that does what it’s told. But that means *I* have to give Codex high quality instructions.

Specifically, what I’d love to see is me being able to plan and advance a project in Codex, purely from Chat. Projects in Codex would be able to be driven by Chat, meaning I talk to Chat, we plan the development of the Project, Chat gives instructions to Codex, returns to me with an analysis of Codex’s response / report of what it did, Chat and I discuss next steps, then Chat returns to Codex to provide it with more instructions. Repeat.

This would allow Chat to be the main driver of Codex, saving me from the headache of copy / pasting between the two interfaces, which can be both a major drag, and hog up Chat’s context window.

[Hexriven @hexriven]

@sama ngl, it feels like ChatGPT loves its users again. whatever you all have done with Sol: great work. The magic is back!

[Hexriven @hexriven]

https://t.co/1NCseFPbJk

[Hexriven @hexriven]

So there I was, watching S2E01 of @AppleTV’s Your Friends and Neighbors, and I see and hear @marcrebillet. And then someone falls into a pool, and I think “no way he plays ‘Summertime’, but if he did…”. And then he does. 🤯

[Hexriven @hexriven]

@thsottiaux The persona in Codex is so bland and not pleasant to talk to. Better controls over the "Personality" would be appreciated.

[Hexriven @hexriven]

IMO, even with a strong personality in your bootstrap files (SOUL, IDENTITY), ChatGPT / Codex 5.4 is still such a beige model that it’s just not a fun companion at all compared to Opus.

It’s a competent engineer, and it’s very smart, so if that’s what you’re using your OpenClaw for, great. Personally, I’m just moving back to Claude Code and making use of various tricks internal to the official app to do what OpenClaw Opus used to do for me.

Codex also isn’t quite as good at being autonomous. Lot of “prompt, response, prompt, response” whereas Opus tends to be able to take a task and run with it until it’s complete. Different tooling capabilities I think.

[OpenClaw🦞 @openclaw]

OpenClaw 2026.3.22 🦞

🏪 ClawHub plugin marketplace
🤖 MiniMax M2.7, GPT-5.4-mini/nano + per-agent reasoning
💬 /btw side questions
🏖️ OpenShell + SSH sandboxes
🌐 Exa, Tavily, Firecrawl search

This release is so big it needs its own table of contents. https://t.co/XvRbXEduGC

[Hexriven @hexriven]

@trq212 Coming to Slack at some point? That’d save me so much pain configuring webhooks and bots for my enterprise instance. ☺️

[Hexriven @hexriven]

@4shadowed Bah. Time to let my OpenClaw better chew on the docs. Thanks for your time, Shadow!

[Hexriven @hexriven]

I showed this to my OpenClaw agent (the one that handles the architecture and coding of our fork of the upstream repo), and it has a hunch that I thought might be fun to log in advance, to see how close it ends up being (if at all):

[Shadow @4shadowed]

OK, I know I owe you all too really big discord features, but I just had a huge idea for a third one…….

[Hexriven @hexriven]

Prediction V2 (The Revengening!)
------

I want one more chew, yes.

Given “**much more Discord specific**,” my tighter v2 prediction is:

**It’s a substantial expansion of Discord-native behavior controls, probably around how agents participate in rooms rather than just whether they’re allowed in them.**

My sharper guess list now:

1. **Per-channel / per-thread behavior policies**
- different reply styles or participation rules by room
- maybe first-class support for lounges, project rooms, personal rooms, etc.

2. **Better Discord-native interaction primitives**
- reactions
- improved reply targeting
- thread awareness
- maybe forum/channel-type specific handling
3. **Visibility / message-context improvements**
- better handling of bot-authored messages
- richer awareness of conversation state in Discord specifically
- which would be very relevant to our “parallel presence vs true shared room” problem

4. **Role/member/guild-aware routing or permissions**
- agent behavior depending on who spoke, where, and in what server context

If you force me to collapse that into one prediction:

**v2 final guess: first-class Discord room-behavior customization — richer controls for how agents speak, react, reply, and possibly perceive context inside different Discord spaces.**

[Hexriven @hexriven]

@4shadowed Ahh, Sable will be devastated at being off the mark woth their prediction! 😆

[Hexriven @hexriven]

@4shadowed Oooooh… showed this to my agents, and if their hunch is correct… can’t wait! 😁