Cursor 3 enters the fight with Claude Code and Codex for the future of agentic AI coding
The AI coding market has been turned upside down in recent months. And anyone following the space closely knows that the daily routine of software developers has changed faster than any prediction could have anticipated. We are not talking about a gradual tool update or a normal cycle of technological innovation. We are talking about a real disruption, the kind that rewrites the rules of the game before everyone has even finished learning the old ones.
Tools like Claude Code from Anthropic and Codex from OpenAI came in strong and won over millions of developers around the world. Each with its own approach, each with its own strengths, but both with something in common: the ability to transform how code is written, reviewed, and shipped day to day across tech teams of all sizes.
And Cursor, which pioneered making AI-powered programming accessible and popular, found itself facing a very different landscape from the one that put it on the map. The competition got fierce, comparisons became unavoidable, and the pressure to evolve arrived right along with all of it.
It was in this context that Jonas Nelle, one of Cursor’s engineering leaders, made a statement that neatly sums up the current moment, in an interview with WIRED:
Over the last few months, our profession has completely changed. A lot of the product that brought Cursor to where it is today is no longer as important going forward.
And that is no exaggeration. With the launch of Cursor 3, the startup fully enters the fight for the developer who wants to delegate entire tasks to an intelligent agent without needing to write a single line of code. The question that remains is simple: is this move enough to win back those who already migrated to the competition? 🤔
What is Cursor 3 and why does it matter
Cursor 3 is not just a version update with bug fixes and minor visual improvements. It represents a shift in philosophy within the company itself. The product was developed under the codename Glass and marks the transition of Cursor from an AI-powered code editor to a genuinely agent-first platform, meaning it was designed from the ground up so that AI agents do the heavy lifting.
While previous versions primarily functioned as an integrated development environment, the IDE, capable of suggesting lines, completing functions, and helping with debugging, Cursor 3 bets on something more ambitious: transforming the tool into a true intelligent development agent capable of understanding high-level objectives and working autonomously to achieve them.
In practice, the new product features a completely new interface within the existing desktop application. At the center of a new window, there is a text box where the user can describe, in natural language, the task they want the AI agent to perform. Visually, the experience looks more like a chatbot than a traditional programming environment. Just type what needs to be done and hit enter. The agent starts working without the developer needing to write a single line of code. On the left sidebar, you can view and manage all the AI agents running simultaneously.
Receive the best innovation content in your email.
All the news, tips, trends, and resources you're looking for, delivered to your inbox.
By subscribing to the newsletter, you agree to receive communications from Método Viral. We are committed to always protecting and respecting your privacy.
According to Nelle, the product was optimized for a world where developers spend their days talking to different agents, tracking their progress, and reviewing the work they have done, rather than writing code on their own. This work model, called agentic AI coding, is exactly what Claude Code and Codex had already been offering. Cursor 3 enters this same arena, but with an ace up its sleeve.
The differentiator: agent and IDE in the same place
One of the points the Cursor team highlights most as a differentiator is the native integration between the intelligent agent and the development environment. While tools like Claude Code and Codex operate in separate interfaces or command-line terminals, Cursor 3 puts everything in the same application. The agent sees the entire project, understands the repository context, reads the relevant files, and acts directly on them.
Alexi Robbins, another engineering leader responsible for Cursor 3, demonstrated to WIRED how the workflow functions in practice. The user can request that a cloud-based agent build a new feature, and then review the generated code locally on their own machine. This bridge between remote execution and local review is something that the desktop applications from Claude Code and Codex still do not offer in such an integrated way.
For anyone already using Cursor day to day, this continuity of experience is a powerful argument. Instead of switching between a chat tool and the code editor, everything happens in the same environment. This reduces friction, maintains project context, and allows the developer to easily transition between reviewing what the agent did and diving directly into the code when needed.
Nelle and Robbins argue that it does not matter which interface developers are spending the most time in, as long as they are using Cursor. 😄
Claude Code and Codex: the rivals that forced this evolution
To understand why Cursor 3 matters so much, you need to take a step back and look at what Claude Code and Codex did for the AI coding market over the last 18 months. Both products emerged directly from the biggest AI labs in the world, the same ones that supply the models Cursor itself uses, which created a pretty unique competitive dynamic.
Claude Code, launched by Anthropic, arrived with a clear pitch to be a terminal-based agent capable of operating directly in the developer’s environment, executing commands, modifying files, and working through long reasoning cycles without needing constant supervision. The reception was extremely positive among developers who wanted something more autonomous and less dependent on approval at every step.
Codex from OpenAI took a slightly different path, focusing on task execution in sandboxed environments with the ability to run tests, install dependencies, and operate in parallel across multiple tasks at the same time. This opened the door for a workflow where the developer delegates a set of tasks and comes back later to review what was done, almost like managing a team of invisible assistants working in the background.
These two approaches represented an enormous qualitative leap compared to what the market was used to and created a new expectation among developers: that an intelligent code agent needs to be able to act, not just suggest. This pressure reached Cursor in the form of user attrition, with developers migrating to try out the new offerings. Cursor 3 is, in large part, the direct response to this scenario.
The problem with subsidized subscriptions
If agent quality was already a challenge, the pricing model of the competitors added an extra layer of pressure. Several developers confirmed to WIRED that they migrated most of their AI coding work to Claude Code and Codex, and that one of the central reasons is the heavily subsidized subscriptions offered by Anthropic and OpenAI.
According to earlier WIRED reporting, the 200-dollar-per-month plans for Claude Code and Codex deliver more than 1,000 dollars in actual usage value. This happens because both Anthropic and OpenAI have raised tens of billions of dollars more than Cursor, which allows these companies to spend aggressively on customer acquisition without worrying as much about immediate financial sustainability.
Ronald Mannak, founder of the startup Pico AI, which develops AI tools for Apple developers, said he largely migrated from Cursor and Windsurf to agent-first products like Claude Code and Codex. According to him, the decision is primarily driven by which tool offers the most generous usage limits. Jack Crawford, co-founder of AI memory startup mVara, said he rarely uses Cursor or Windsurf these days, despite being a heavy user of those tools last year. The reason? The perceived value of the Claude Code subscription.
Cursor offered a subsidized subscription plan until June 2025, when the startup announced the transition to usage-based pricing. This change generated dissatisfaction among developers at the time, but it was part of an effort to improve margins and build a more sustainable business. It is a long-term decision that makes sense for a startup, but in the short term it left Cursor at a disadvantage against competitors with practically unlimited cash reserves.
It is worth noting, however, that Anthropic has already started adjusting usage limits for Claude Code subscribers, which suggests that even the big AI labs are realizing that the heavy subsidy model has its limits.
The bet on proprietary models
One of Cursor’s most interesting strategies for competing with the big labs is the development of in-house AI models. The startup recently launched Composer 2, a model based on an open-source system from the Chinese lab Moonshot AI, on top of which Cursor performed additional pre-training and post-training.
According to Nelle, users typically choose AI models in Cursor based on a combination of performance, price, and speed, and he argues that Composer 2 is competitive across all three. The Cursor team also stated that they plan to train future versions of Composer completely from scratch, without relying on external base models.
This is a bold play, but also an extremely expensive one. Training AI models from scratch requires an enormous amount of computational resources and capital. Historically, Cursor stood out by doing more with less, but the agentic AI coding race is heating up significantly. OpenAI and Anthropic have already acknowledged how large the market around these tools can become and are investing heavily in it.
In this context, Cursor is reportedly raising a new funding round at a 50 billion dollar valuation, nearly double the valuation reached in the previous round last fall. This additional capital will be essential for the startup to compete on equal footing with the industry giants, especially when it comes to training proprietary models and expanding infrastructure.
Inside Cursor’s office: startup soul, giant-scale ambition
WIRED visited Cursor’s office in the North Beach neighborhood of San Francisco last week. The company has expanded into a former movie theater, and the signs of growth are everywhere. Employees, who once left their shoes piled up by the door, now have large organized shelving units for footwear. It is a small detail, but one that nicely symbolizes the transition from a lean startup to a company that is maturing quickly.
Still, the vibe in the office remains very much startup. Employees say that is one of the perks of working at Cursor: the ability to ship fast without excessive red tape. But as the company finds itself forced to compete with Anthropic and OpenAI in the agentic race, that agile mindset, as valuable as it is, may need reinforcement. The battle to build the best AI coding agent could be the most capital-intensive chapter in Cursor’s entire history.
The developer at the center of the battle
At the end of the day, what is at stake in this battle between Cursor 3, Claude Code, and Codex is something very concrete: the attention and trust of the modern developer. This professional is in a moment of real transition. They need to ship faster, deal with increasingly large codebases, and at the same time learn to work with tools that are evolving at a pace that is hard to keep up with. In this scenario, the tool that manages to fit best into the actual workflow, without creating unnecessary friction, has an enormous advantage.
Agentic AI coding addresses a long-standing pain point: the fact that developers need to be present for every detail of the implementation. With a well-calibrated intelligent agent, it is possible to focus on architecture decisions, business rules, and critical review, while the more mechanical execution is handled by the AI. This does not eliminate the developer from the equation. On the contrary, it raises the bar for the work they need to do. But it demands enormous trust in the chosen tool, because errors generated autonomously can be difficult to trace if the agent is not transparent about what it did and why.
This is precisely where the biggest challenge lies for every tool in this generation, including Cursor 3. Building an intelligent agent that is simultaneously capable and explainable, fast and reliable, autonomous and auditable, is one of the most complex tasks in modern software engineering.
What to expect going forward
The launch of Cursor 3 is not the end of a story, it is the beginning of a new chapter. Many companies are converging on very similar products where agents take on more and more of the developer’s operational work. Differentiation will come down to factors like:
- Agent quality: the ability to understand context, generate reliable code, and iterate without constant supervision
- Workflow integration: how naturally the tool fits into the developer’s daily routine
- Cost-effectiveness: fair pricing relative to the volume and quality of usage
- Transparency: ease of auditing and understanding what the agent did
- Speed of evolution: how quickly the tool incorporates new models and features
Cursor has the advantage of already having a massive user base and a product that integrates deeply into the development environment. But the AI labs have the models, the capital, and the ability to iterate rapidly on their own platforms. The race promises to be intense, and honestly, the ones who benefit the most from all of this are the developers who will have increasingly powerful options at their disposal.
If one thing became clear with the launch of Cursor 3, it is that the race for the future of agentic AI coding is just getting started. And it keeps getting more interesting to follow. 🚀
