Great question. I've read that the agent sdk is ok as long as it's not for external use - meaning you aren't selling access to it. Unclear now though!

Big Giant Million Dollar Question: Where does having Openclaw using Claude Code via ACP fall? It's using the Claude Code harness, not the model directly.

If you are not aware, ACP creates a persistent session for steering rather than using the models directly.

The Zed ACP client for example is still controlled by the human prompt, and they will probably not be banned

I have no idea what ACP offers that are superior to a tmux session. With tmux, you can attach to it at any time, send keys at any time, and capture pane without bothering any running process inside.

And you don't have to get anyone's permission to use tmux.

> And you don't have to get anyone's permission to use tmux.

I'm not so sure about that. I have my own local multi agent orchestration setup with tmux and native claude code, codex, and gemini. They can talk between each other using tmux.

I'm not sure precisely when does a wrapper around Claude Code become a "third party harness". If OpenCode is a third party harness, why is tmux not?

on spirit, yes. programmatically speaking, not really.

And judging by intention is a slippery slope, and they'll probably end up just banning everyone who are high usage. While some smart orchestration setup will fly under the radar (i.e. trivial to rate limit the cross tmux communication).

This is my big question too. It seems by intent it's to kill it, including ACP, but I don't know.

Started using this earlier this week. I built a backtesting benchmark tool to compare a mix of frontier and open-source models on a fairly heavy data analysis workflow I’d been running in the cloud.

The task is basically predicting pricing and costs.

Apple’s model came out on top—best accuracy in 6 out of 10 cases in the backtest. That surprised me.

It also looks like it might be fast enough to take over the whole job. If I ran this on Sonnet, we’re talking thousands per month. With DeepSeek, it’s more like hundreds.

So far, the other local models I’ve tried on my 64GB M4 Max Studio haven’t been viable - either far too slow or not accurate enough. That said, I haven’t tested a huge range yet.

this is great! Incredibly fast and is working pretty well running loads on my m4 max studio.

Weirdly though I'm getting things like this: Apple FM is fast and free but has a hard limitation — it can't process prompts with Spanish/non-English words, which is a dealbreaker for California and Southwest real estate where half the street names are Spanish.

My understanding is that the 512gb offering will likely return with the new M5 Ultra coming around WWDC in June. Fingers crossed anyway!

I've read this and the replies and I don't understand - but want to. I have 6 agents spun up under one openclaw. 4 of them use my claude code max sub. 2 use my codex sub. I am not needing to use any metered API calls. Why are you?

I just don’t want to risk an account ban like I’ve heard others describe. But if it’s working for you that’s awesome! Saves a lot of money! I don’t mind the cost personally for the value.

Though you do have a point that OpenAI said their subscription is usable in third party apps so I should be using it then. Good point.

Ah, the way I read your previous comment was that you were using both your plan and the api. I understand now.

Oh no, not at all. But your comment made me give the bot a ChatGPT subscription that it's using with the Codex endpoints since they've announced that they will be supporting 3rd party use of those endpoints. It's a bit slower than Sonnet through OpenRouter but it's vastly cheaper. Much appreciated for the advice!

Potentially amazing opportunity for OpenAI to more meaningfully compete with Claude Code at the developer and hobbyist level. Based on vibes it sure seemed like Claude Code / Opus 4.6 was running away with developer mindshare.

Peter single handedly got many of us taking Codex more seriously, at least that's my impression from the conversations I had. Openclaw has gotten more attention over the past 2 weeks than anything else I can think of.

Depending on how this goes, this could be to OpenAI what Instagram was to Facebook. FB bought Instagram for $1 billion and now estimated to be worth 100's of billies.

Total speculation based on just about zero information. :)

> Peter single handedly got many of us taking Codex more seriously, at least that's my impression from the conversations I had.

Comments like this feel confusing because I didn't have any association between Codex and OpenClaw before reading your comment.

Codex was also seeing a lot of usage before OpenClaw.

The whole OpenClaw hype bubble feels like there's a world of social media that I wasn't tapped into last month that OpenClaw capitalized on with unparalleled precision. There are many other agent frameworks out there, but OpenClaw hit all the right notes to trigger the hype machine in a way that others did not. Now OpenClaw and its author are being attributed for so many other things that it's hard for me to understand how this one person inserted himself into the center of this media zeitgeist

He's been on a number of podcasts - lex recently, and is really emphatic about Codex as the breakthrough solution he relies on. I just looked and on the handful of podcasts there are about 2,000,000 views this past week and half or so.

yes, i switched to Codex after he mentioned it on "Pragmatic engineer" podcast and I ran out of Claude credits on 20x plan. So far Codex is matching or slightly beating Claude Code for me. Loving the Desktop app despite the slowness

It's how Steve Yegge became a "father of agentic orchestration" or something - there is some Canonical Universe Building exercise somewhere on twitter that just looks, for the lack of a better word, not rigorous. But good for all these people, I guess, for riding the hype to glory.

you didn't see because you dont follow peter on twitter. he talked for months now how codex is a better coder.

I’m not disputing that people who follow Peter are getting information from Peter. It’s the “single handedly” part of the claim that was strange.

I’m questioning how some people in that bubble came to believe he was at the center of that universe. He wasn’t the only person talking about the differences between Codex or Claude. Most of the LLM people I follow had their own thoughts and preferences that they advertised too.

Sure, single-handedly is doing a lot of work here. :) Anecdotally a fair number of people I know have referenced his thoughts so I just ran with that. Most people seem to kind of equivocate about whatever model they like, Peter on the other hand is very strident about it.

Superpowers is from Obra (Jesse Vincent), quintessential hacker and was a leader in the Perl community back in the day (still?).

https://github.com/obra/superpowers

The software industry moat is disappearing. What will happen is 1000 (10,000?) competitors putting price pressure on every saas app.

Software will cease to be a winner take all and be a very long tail distribution!

Where are all these 1000 competitors coming from?

You need to have tremendous agency/will to start competing with a public company. Plus, you need to have a lot of distribution channels, competing with sales people that do this for a living since ages, and marketing budgets that are higher than your annual Claude spending.

Regular people do not press 10 clicks daily to track their calories, and you're saying that they will start competing with Salesforce and the like?

GM Supercruise on my 2024 Silverado RST is a joke compared to Tesla FSD. It's not even remotely comparable. Supercruise only works on freeways/highways, does not understand ANY navigation. It's a better cruise control, that's about it. I own 2 Tesla model S of different vintages and FSD is a completely different animal. My 2017 model s can navigate from my house to, well, anywhere, with no intervention. I have been very disappointed in how long it took Tesla to get here based on the promises they made 10(!) years ago, but they are there now. Even a year ago FSD used to scare me frequently and cause me to disengage but that never happens now.

The only thing remotely comparable in the US is probably a Comma 4 with a forked OpenPilot build on a well-supported vehicle