Mark Kretschmann

By the end of this year, AI coding models will be better than the vast majority of human coders. Mark my words. 🔮

@Grummz It depends on the task. If I ask a complex question that requires research, I'm irritated if the model gives an instant answer. That's because I know that the task requires reasoning time.

Every AI company should be optimizing for speed of inference. It’s the difference between at 10fps game and a 60fps game. The same game feels horrible at 10 and amazing at 60. It’s not just a quality be degrees situation, it’s transformative. The same model feels 100x smarter and can get more done with realtime inference.

@osarood @elonmusk @cursor_ai @xai xAI is currently training 3 Grok Build models at the same time.

@elonmusk @mark_k @cursor_ai Will @xai be one of them?

Tomorrow, @cursor_ai is going to release a super strong coding model. Possibly SOTA. Stay tuned. 🫡

@diegocabezas01 @LearnAI_MJ In many cases, yes.

@mark_k @LearnAI_MJ If they aren’t already

By the end of this year, AI coding models will be better than the vast majority of human coders. Mark my words. 🔮

@SKroWL1776 @elonmusk @cursor_ai Grok Build is the name

@elonmusk @mark_k @cursor_ai Can't wait for Grok CLI (that works with my X premium not greedy pay per use API keys)

@savaerx Especially at 3 am 😏

@mark_k for isolated tasks, probably already true. 200k line legacy codebase, no docs, on-call at 3am? that part's still a few model generations out.

@bjornebark The trajectory is really crazy now

@mark_k I agree, did not think that a couple of months ago but yeah correct.

@foley2k2 How many human coders can build that, and how long does it take them?

@mark_k Ok, let's see them build an operating system, low latency networking firmware, or a real-time procedural rendering engine.

@AdemVessell I'm Mark, nothing can go wrong 😉

@mark_k MARKED

@ashcotXBT AI is still very bad with ideation. That task remains with humans for now.

@mark_k the median coder maybe. the taste bottleneck still looks very human

@KuittinenPetri Yes, I agree, at least for certain tasks.

I would already argue that a local model with good AGENTS.md is already better than most human coders. And you don't need a huge model e.g. FP8 quant of Qwen3.5-27B will fit about 32 GB VRAM. Most human programmers are incredibly bad when to comes to comprehensive error checking, fault tolerance and protecting your code base against most common vulnerabilities.

@D2Chattoway Uh oh 😉

@mark_k Noted 😊

@JustinGorya All of the big AI labs have made coding the number one priority.

I think it's already or almost there. But the end of this year will be extremely wild. My personal opinion is that we're seeing prices increase during SaaS right now because sellers know time is running out. Nobody will use SaaS in two years. This is my prediction. They're trying to squeeze the lemon as much as possible while they still can.

@amarinnovates Not for long. Agents will orchestrate agents.

@mark_k AI might win the syntax battle this year, but the devs who evolve into system orchestrators will still hold the high ground.

@Cyn_Cyb3r071Qu3 @Google Real. It's a reply.

@mark_k @Google Can't find that post anywhere. If this is true, we're fucked. Lol

Elon Musk thinks @Google will win the AI race overall! However, he also sees China winning “on Earth” and SpaceX in space with its upcoming AI satellites.

@ptrckwnsch @digitalfoundry More than silly. They're going to destroy our future.

@mark_k @digitalfoundry I'm losing my patience with the anti-AI folks. It's so silly.

Digital Foundry received death threats over their coverage of DLSS 5. 💀 "We’ve received death threats… this is completely unacceptable. No one should face threats for doing their job or sharing tech." (@digitalfoundry) Fanatical decels hate DLSS 5 because it is a form of AI.

@DivineRational @cursor_ai @nvidia Soon new coding model

@mark_k @cursor_ai @nvidia Yeah. I have a feeling something big is coming.

Don't count Cursor out just yet. They might be back, big time. @cursor_ai 🤝 @nvidia

@priestessofdada @Google That's what he wrote

@mark_k @Google Nah. China's winning the AI race with their open weight models. Deepseek, Qwen and Kimi are being deployed faster and more often at the small to mid-enterprise level. They're also integrating faster into new AI products. American AI is wildly misunderstanding the play

@YuriKushch @cursor_ai I hope so too

@mark_k @cursor_ai Hopefully that quality and quotas will be a lot better or comparable with what we have now in Codex and Claude Code