“Microsoft has done a number of fantastic things, but I don’t think they’re in a great place to really keep innovating and pushing on this in the way that a startup can,” said Aman Sanger, one of the creators of Cursor (by Anysphere), in a recent episode of the Lex Fridman podcast. When asked how Cursor, a newbie, could compete with a well-established Microsoft product like VS Code with Copilot, he said that the tech giant lacked the research and experimentation necessary to really push the ceiling.
“I don’t know how you put that into words, but when you compare a Cursor with a Copilot, Copilot pretty quickly starts to feel stale for some reason,” resonated Fridman.
Citing OpenAI’s o1 and others, Anysphere co-founder Sualeh Asif said when they started Cursor, they really felt this frustration that you could see models getting better, but the Copilot experience had not changed. “It was like ‘Man, the ceiling is getting higher, why are they not making new things?’ They should be making new things,” he added, “Where’s all the features? There were no alpha features.”
He said it was selling well and surely did great business, but he is one of those who really want to try and use new things. “And there were no new things for a very long time,” added Asif.
Another co-founder of Anysphere, Arvid Lunnemark said Cursor is like an all-in-one platform for developers, touching upon the proximity and often overlap of responsibilities within the teams—where the person designing the UI/UX is also involved in training the model—allowing for seamless integration and experimentation. This approach makes it possible to create things that wouldn’t be achievable without constant communication and collaboration.
“And you’re using, like you said, Cursor to write Cursor?” said Fridman, adding that the team is rapidly implementing features, alongside doing the research experimentation necessary to really push the ceiling.
Founded by Truell, Asif, Lunnemark, and Sanger, Anysphere’s Cursor started with the goal of writing the world’s software. The founding team said they were fans of VSCode and Microsoft Copilot, and getting early access to GPT-4o served as their path and foundation to building Cursor.
The rest is history.
Microsoft Integrates o1
A few days ago, GitHub made OpenAI’s o1 model available on Copilot. This integration was key to competing with the likes of Cursor.
“So it looks like GitHub Copilot might be integrating o1 in some way, and I think some of the comments are saying, does this mean Cursor is done?” asked Fridman. “It’s time to shut down Cursor, yeah,” retorted Lunnemark.
The Cursor team spoke about how they are experimenting with integrating the o1 model into the Cursor editor as well. And while there is curiosity from programmers, it is still not a default experience for coders as a clear use case hasn’t emerged yet.
“I think most of the additional value from Cursor versus everything else out there is not just integrating the new model fast like o1. It comes from all of the depth that goes into these custom models,” said Sanger, stressing how thoughtful design and user experience is more important.
Conversely, research shows that OpenAI’s o1 models, known for strong reasoning abilities, struggle with coding tasks, leading to slow responses, hallucinations, and reliance on more effective models like GPT-4 and Claude Sonnet 3.5 for code generation, sparking widespread developer dissatisfaction.
The Rise of Cursor Alternatives
Earlier this year, Zed AI, a code editor which brings LLMs directly into your editor with a text-centric approach, was launched in collaboration with Anthropic. Zed was founded in 2021 by Nathan Sobo, who previously worked at GitHub, where he built and led their Atom text editor team.
Additionally, two coding assistants that have developed their own foundation models, Codeium AI and Magic AI, founded by Itamar Friedman and Eric Steinberger & Sebastian De Ro, respectively, reached unicorn status in September following significant funding rounds from investors such as General Catalyst, Sequoia, and Nat Friedman and Daniel Gross.
Recently, OpenAI introduced Canvas, a new interface for writing and coding that expands ChatGPT’s functionality beyond simple conversation. Anthropic’s Claude Artifacts, which is available to all users on iOS and Android, allows individuals to easily create apps without writing a single line of code.
Meanwhile, Y Combinator has also been backing several AI code editors, especially since the launch of Cursor AI. They have funded startups like Continue, Pearl, Void, Type, and Melty, among others, which are direct open-source competitors to industry incumbents like Cursor AI and Zed AI, among others.
“Fast is Fun”
Needless to say, developers are increasingly seeking AI tools that help them code efficiently by automating repetitive tasks and providing real-time predictive assistance. A GitHub survey showed that 97% of developers had used AI coding tools at work, regardless of employer approval. The trend is going to continue.
With Cursor’s Tab feature, developers can eliminate ‘low entropy’ actions, such as typing predictable code snippets, enabling faster workflows. The future of programming will see “all of programming flow through these models”, said Asif, indicating a major shift in how developers build software.
With Cursor focusing on speculative edits and caching to deliver real-time assistance, speed is a crucial factor. According to Senger, Speculative edits allow the system to “predict ahead” of the developer’s actions, streamlining the coding process to near-instant responses.
Cursor claims to make current processes obsolete within a year, reflecting how developers prioritise tools that accelerate workflows while maintaining high code quality.
“Even being a few months ahead in AI programming makes your product much, much more useful,” said Truell, adding that Cursor offers developers a faster coding environment using sparse models, speculative decoding, and multi-query attention, thereby significantly reducing latency.
“I actually think this is a really, really exciting time to be building software,” he said, reminiscing about the good old programming days of 2012-2013.
“So, it’s going to be a really, really fun time for people who build software and the skill will probably change too. I also think that people’s taste and creative ideas will be magnified,” he added, saying, “Fast is Fun.”
“That should be a T-shirt,” quipped Fridman.
The post ‘Microsoft isn’t in a Great Place to Really Keep Innovating’ appeared first on AIM.
