Enjoying this issue?
Get tomorrow's AI & engineering digest in your inbox — hand-picked, summarized, and always spam-free.
TLDR
Grok 4.7 is a fast, cheap model close to Fable 5.1 but not Astra-level. The recommended workflow uses Astra to create a plan, then launches a sub-agent with Grok 4.7 for execution. Cloud Room's GUI now clearly distinguishes local and cloud threads and allows switching reasoning efforts via text commands.
Key points
Grok 4.7 is an improvement over 4.6, close to Fable 5.1 but faster and cheaper.
The recommended workflow uses Astra to create a plan, then Grok 4.7 as a sub-agent.
Cloud Room GUI now clearly shows which threads are local and which are cloud.
Users can switch reasoning efforts by typing commands like 'X high' in the prompt.
The UI for selecting local or cloud thread startup is now clearer.
Tools mentioned
Techniques
- plan-then-execute workflow using two models (Astra for planning, Grok 4.7 for execution)
- switching reasoning efforts via text commands in the prompt
- local vs cloud thread management
Stop scrolling. Start reading smarter.
Receive the day's most important AI & engineering updates in one concise email. No spam.
Transcript (captions)
recording. All right, so Grog 4.7 is here. It just dropped just straight up improvement over Grog 4.6. It's not going to be Astra level, right? But seems to be close to Fable 5.1.
Obviously way faster and way cheaper. So instead of cloud room if you want to use it, there's obviously a cursor CLI adapter. I mean the Cloud Room GUI is fork of BB. So here you can just do Gro
47. Boom. I would prefer like default with fast high fast extra high if you have loads of credits or if you don't care. But yeah, this is a great model for you know everyday work. I mean if
you need frontier intelligence obviously use Astra but the way to use it is actually have Astra create a plan. So like I want to implement feature XY Z create a plan for it right and then you
can u just launch a sub agent in the cloud. So for example now we can say like use uh cloud room CLI to launch a sub agent on my uh David VM cloud and have him investigate
whether we have the plan for this feature right so the Astra will just write the plan and then uh we can have a gro 4.7 and I I can actually interrupt it and change say boom use grog 4.7
as the model for the sub aent via P by um harness. So you can just edit messages and uh Astra will launch it as a sub agent. And here actually new thing is that you can
very nicely see which threads are local and which are cloud. So obviously for some workloads it still makes sense to run a local agent like if you want to clean up you know some buggy install or
you know fix some settings on your computer or maybe your downloads folder has too many files you can just launch a local thread but also I fixed the UI. So you can see here selecting whether to
start the local or cloud thread is very clear now. So local either primary checkout or in a work tree or cloud primary and then we'll add worksheet support later. So now it's very clear if
you're selecting local or cloud. And also another nice thing is that you can just easily switch reasoning efforts. So you're just typing in like blah blah blah. You can just do X high and u you
know switch to extra high or medium reasoning efforts. And also it works on the cloud now because last time it was kind of bugged. So you can just do like low. I switched to reasoning effort low
which is very nice if you're like typing a prompt and just want to switch to extra high. It's so much faster than having to like click click click, you know. So, we're making both the GUI as
smooth as possible, as easy to use as possible, while also making uh the core support more harnesses and be more stable and yeah, just making cloud room better in every way.
>> Looks clean.