Arthur Katcher

CTO & co-Founder @QORIS_AI • building the secure harness for enterprise AI agents // 1x Acquired • Claude Certified Architect

[Arthur Katcher @arthurkatcher]

You can now run Claude Code & Codex right from Grok Build!

I've just open-sourced a plugin to make it possible.

With Grok-Delegate you can use Grok 4.5 as your main fast orchestrator and delegate hard tasks to models like Fable 5 or GPT 5.6 Sol.

Just install the plugin and run /delegate-claude or /delegate-codex.

If you're seriously considering using Grok Build as your only harness, as I am, but can't do that because of the need for the big models, this plugin can really help.

The link is in the comments ⬇️

[Andrew Milich @milichab]

@arthurkatcher 1) Yes! Use `.grok/agents` or config.toml (see below), adding to the docs site right now.

2) Yes: Agents can spawn background subagents, or try /dashboard, to move your agent to the background and create another. I've been using /dashboard as my default.

[Elon Musk @elonmusk]

Using Grok agents in background mode

[Arthur Katcher @arthurkatcher]

@adityawaslost It's actually good, plus the UI/UX is really enjoyable

[Arthur Katcher @arthurkatcher]

You can now run Claude Code & Codex right from Grok Build!

I've just open-sourced a plugin to make it possible.

With Grok-Delegate you can use Grok 4.5 as your main fast orchestrator and delegate hard tasks to models like Fable 5 or GPT 5.6 Sol.

Just install the plugin and run /delegate-claude or /delegate-codex.

If you're seriously considering using Grok Build as your only harness, as I am, but can't do that because of the need for the big models, this plugin can really help.

The link is in the comments ⬇️

[Andrew Milich @milichab]

@arthurkatcher 1) Yes! Use `.grok/agents` or config.toml (see below), adding to the docs site right now.

2) Yes: Agents can spawn background subagents, or try /dashboard, to move your agent to the background and create another. I've been using /dashboard as my default.

[Arthur Katcher @arthurkatcher]

@ZaharovAndrew For sure, why not. Maybe will add as a new repo in a couple of days.

[Arthur Katcher @arthurkatcher]

@RonfortMartin Yeah, we need to grow the infrastructure around Grok Build. Plugins from the public sector can really help with that.

[Arthur Katcher @arthurkatcher]

https://t.co/jVPLN5kRkS

[Arthur Katcher @arthurkatcher]

@adelbucetta I guess we will see!

[Arthur Katcher @arthurkatcher]

Grok Build is an awesome harness with a set of distinct and pretty unique features.

The UI/UX is great, one of the best out there, actually.

Elon and xAI aren’t kidding; they’re emerging as a solid player. If they scale their models right, Grok Build might take up more and more space in my development lifecycle.

It’s still missing some core features, though. You can’t background an agent yet, things like that.

But it’s a solid B+, maybe even A. I’ll keep an eye on Grok Build for sure.

[Arthur Katcher @arthurkatcher]

Built fully with Grok Build & Grok 4.5 - and I have to say, it is solid, very solid.

Not Fable 5. More like Sonnet 5, but faster.

A good model for basic discovery, but I still would not trust it with the critical flows 100%

[Arthur Katcher @arthurkatcher]

@ApplyWiseAi For sure

[Arthur Katcher @arthurkatcher]

@theo Claude Code is just a better harness than Codex, it scores higher on all the benchmarks

[Arthur Katcher @arthurkatcher]

@ApplyWiseAi 100%

[Arthur Katcher @arthurkatcher]

This is the comparison of the Plan modes.

Grok build on the left.
Claude Code on the right.

In Claude Code you can only give 1 overall comment to refine the plan, while in Grok Build you can comment per particular line.

Clearly Grok Build wins this one.

[Arthur Katcher @arthurkatcher]

One thing I can't get enough of Grok Build - the comment feature in the plan review mode.

You can just type your own comment on every line of the plan, and Grok will adjust it accordingly with the context.

This is literally UX gold. Claude Code should implement this.

[Arthur Katcher @arthurkatcher]

Grok Build xAI team is there to help you and listen to the feedback.

Unbelievable, Claude Code & Codex should watch out.

Grok 5 it, and I will use only Grok Build and nothing else 😀

[Andrew Milich @milichab]

@arthurkatcher 1) Yes! Use `.grok/agents` or config.toml (see below), adding to the docs site right now.

2) Yes: Agents can spawn background subagents, or try /dashboard, to move your agent to the background and create another. I've been using /dashboard as my default.

[Arthur Katcher @arthurkatcher]

@milichab Awesome, thank you!! Missed with the second one

[Arthur Katcher @arthurkatcher]

Been using Grok Build with Grok-4.5 for a couple of days -

and it's actually a solid harness. You can build complex projects with it end-to-end.

Still a long way to go to reach Claude Code, but surprisingly good.

[Arthur Katcher @arthurkatcher]

One thing I can't get enough of Grok Build - the comment feature in the plan review mode.

You can just type your own comment on every line of the plan, and Grok will adjust it accordingly with the context.

This is literally UX gold. Claude Code should implement this.

[Arthur Katcher @arthurkatcher]

@mrlesk 100%, let's make that the standard

[Arthur Katcher @arthurkatcher]

@milichab Hey, thanks for writing! A couple of things I've noticed so far:

1. a possibility to create your custom subagents. or is it there, and I'm missing it?

2. backgrounding an agent

Those are the two things that I rely on heavily in the daily work

[Arthur Katcher @arthurkatcher]

Built fully with Grok Build & Grok 4.5 - and I have to say, it is solid, very solid.

Not Fable 5. More like Sonnet 5, but faster.

A good model for basic discovery, but I still would not trust it with the critical flows 100%

[Arthur Katcher @arthurkatcher]

Gave Grok 4.5 a test run.

Everyone knows MCPs are a pain, they flood the context window, you have to load them right, manage tokens… yikes.

The fix: one MCP to rule them all.

Wire your many servers into a single connection. Expose just one endpoint with just 8 tools to Claude, ChatGPT, Cursor, or any agent.

Fully open source, self-hosted. OAuth both ways: agents log into MetaMCP, MetaMCP logs into protected upstreams.

Save tokens. 1 MCP instead of 100.

Link in the comments.