How Cursor is evolving through its Composer coding models built on Chinese open models, as coding agents like Claude Code threaten to make code editors obsolete
After becoming the hottest, fastest growing AI coding company, Cursor is confronting a new reality: developers may no longer need a code editor at all.
Forbes
Context & Ripple Effects
Cursor’s rapid growth has put its product boundary under pressure: the relevant competitive question is no longer only which editor offers the best AI assistance, but whether the editor remains the primary place developers direct coding work.
Cursor must evolve Composer from an editor enhancement into a capability that can compete with standalone coding agents, while retaining developers who may otherwise move their work outside an editor.
Using Chinese open models makes the performance and integration of Cursor’s own Composer models a more immediate part of its product differentiation.
Second-order effects
The competitive benchmark for AI coding products shifts toward autonomous task execution and agent management, rather than inline completion or editor features alone.
Rivals such as Claude Code raise the cost of being a single-agent coding interface; Cursor’s later multi-agent product direction indicates that orchestration is becoming a necessary response.
Third-order effects
If agent-driven workflows continue to displace editor-centric ones, durable value may accrue to the work surface that holds project context and coordinates agents, not necessarily to the traditional editor itself.
The market could separate model providers from workflow owners: open-model-based products can compete through coding specialization and orchestration, but their defensibility will depend on whether those layers remain differentiated.
The trend: AI coding is moving from editor-embedded assistance toward agentic work surfaces that execute and coordinate larger portions of software development.
Good piece on the “war time” at Cursor. Some interesting quotes: - The company's new mandate was labeled “P0 #1”—priority zero: “Build the best coding model.” - Cursor estimated last year that a $200-per-month Claude Code subscription could use up to $2,000 in compute, suggesti…
Cursor internal analysis shows how hard Anthropic is subsidizing Claude Code. Last year, a $200 monthly subscription could use $2,000 in compute. Now, the same $200 monthly plan can consume $5,000 in compute (2.5x increase). [image]
Another huge priority is enterprise: though it reached $100m ARR with no sales staff, now half the company's staff is in sales. Traditionally, Cursor's enterprise revenue has been a small portion of its total revenue (which is still growing and passed $2b ARR recently). [image]
The company believes the real reason behind many developers' switch to Claude Code is Anthropic's aggressive subsidization: a $200 Claude Code individual plan can actually utilize more than $5,000 in compute, according to some estimates.
It's so weird how for Anthropic the source has such specifics but Cursor has this wishy washy “profitable margin” business product. Someone I was talking to with a cursor enterprise account who was able to spend $20k+ in credits on a $9k contract, not sure how those margins work
Cursor is now trying to be the best at orchestrating hundreds of agents working together, internally called ‘Grind mode.’ Coding is going through massive changes, and only this is certain: the AI coding wars are only just beginning. I'd say, don't count @cursor_ai out!
There has been so much chatter about @cursor_ai and whether Claude Code is killing it, so we did reporting to see if that's true. Here are some highlights, and read our story to go behind-the-scenes at what's going on at Cursor: https://www.forbes.com/... @RashiShrivast18
I ran the numbers on my day yesterday: 10 hour day of coding, no meetings. It was $118 worth of tokens. In one day. I've heard of many devs burning through $350 worth in a day. And this is at subsidized prices
We saw this happen with Uber and Lyft 10 years ago and now it's a duopoly that's bad for the drivers and the riders I don't see how heavily subsidized inference ends well, inference providers drop the costs now build moats and then jack it up later
when you see articles like this you have to assume this information was deliberately given to a journalist and ask why - suggests claude code's success comes from subsidization (cursor didn't lose) - they are going to make the best coding model (don't churn you'll be back soon)
Michael or someone else at Cursor needs to hit the podcast circuit. This narrative feels forced. Cursor is at a point growth wise where a coding model makes sense as a P0. LLMs were always going to get here, they just needed a detour through vscode first.
Internally, Cursor see itself as a frontier model company in “war time”, with its primary goal being building the best coding model. But Cursor is *highly* dependent on the Chinese open-source ecosystem, because their own models are based on DeepSeek, Qwen, etc. [image]
After becoming the hottest, fastest growing AI coding company, Cursor is confronting a new reality: with the rise of powerful AI agents developers may no longer need a code editor at all. With @annatonger. https://www.forbes.com/...