X No. 14912 words
Why I'm so excited about Cursor and Grok Build, even though Codex + GPT 5.5 writes better code today

I’ve been working with Grok Build since the day after it was released. Every single day I’ve been doing my best to really push it to its limits, and learn.
What I think makes Grok Build so important is that it will be powered by Grok 5 at some point this year, and when that happens, it could become the most powerful agentic coding platform on the planet.
This is why you see me post so much about Grok Build, when, today, in practice, me and my team are using Codex w/GPT 5.5 as our daily driver.
Yes, today Codex with GPT 5.5 does, without a doubt, produce better production-grade code. And as we saw with DeepSWE, the first benchmark to truly test with real world agentic coding tests - GPT 5.5 is better at coding that Opus 4.7.
But, and this is a big but. I believe that if you’re in the agentic coding world, you need to be good at two things:
- The platform + model that is the best at coding today - and that is Codex + GPT 5.5, no question.
- The platform that has a high likelihood of being the best at coding tomorrow - I think this could be Cursor/Grok Build + Grok 5.
If you use Grok Build today, you’ll find lots of sharp edges, and nope, it won’t produce code as good as Codex w/GPT 5.5, and yup, it’ll ship bugs.
At the same time, I’d argue that part of this is on you. Codex allows you to be lazy in some ways, it has an incredibly well-crafted way of working, so you can give it a pretty simple prompt, and it will produce good code.
Grok Build doesn’t work like this today. After two weeks of testing, and comparing side-by-side with Codex, I can tell you - if you give Grok Build the same prompt as Codex, it might only build half of what you tell it to, and it could have some pretty gnarly bugs.
But what’s important to remember here is that Grok Build is playing catch up, this is an early beta, and they aren’t claiming to be the best agentic coding tool on the market. I think the SpaceXAI team has been very open and transparent about this.
What they have said is - this is an early beta, give us feedback, it will get better. And that’s exactly what has been happening, daily. OpenAI has Codex Thursday’s where they ship new stuff, the Grok Build team ships, and shares what they ship, every single day.
This morning, or late last night, depending on when you go to bed, Elon tweeted this:
SpaceX has almost finished writing V1.0 of an in-house AI training stack in C that exact-maps to 220k GB300s with 800G NICs, making heavy use of pipeline parallelism and getting as close to bare metal as possible.
The potential speed improvement vs JAX for large training runs is
And someone then asked him if it would be used for Grok 5, and he confirmed, it will be.

To me, and coupled with the fact that buying Cursor is on the table, I see a very interesting potential future. While a lot of people seemed to forget about Cursor, they have a very powerful platform, one that is used by millions of developers around the world - it’s polished, feature-rich, and Composer 2.5 is a very good model.
If SpaceXAI acquires Cursor, and trains Grok 5 to be the beast of a model that it looks like it’s going to be, we’re going to have a platform/model combo that will be incredibly hard to ignore.
But we aren’t living in that world yet. And so most people don’t really need to do what I’m doing. You can just jam with Codex and GPT 5.5 and be happy.
I wish I could wire my brain this way, but I can’t. I want to live in the future, and prepare myself for what’s next. And while, as a CTO, I haven’t deployed Grok Build our to our team, and I likely won’t for a while here, I want to use and learn Cursor and Grok Build as much as I can, so if this future, that I see coming at us fast, happens, we’ll be ahead of the curve
It’s a kinda annoying thing about me, something that honestly bugs me about myself. I can’t be happy with what currently is, I’m obsessed with what will be, and how I can give myself as much of an edge on what I think will happen. I can’t change this about myself, so I will always be using things that are way too early, have lots of sharp edges, and aren’t as good or polished as something that exists today.
At the same time, it’s really exciting, it’s the future, and while nobody can predict what will happen, it’s fun to try, and in the process, go deep with technologies that are still being built.
So if you see me talking about Cursor and Grok Build a lot, don’t get confused, I’m not a freakin’ content creator, I’m a founder, a CTO, and an engineer that likes to try new shit, broken shit, shit that’s not quite finished yet, and I love playing with it, finding those sharp edges, and learning how to get ahead if the world changes in the ways that I think it will.
Okay, morning rant over - time to play with Grok Build, in Cursor, some more.