Ok so Cursor dropped Cursor 3 yesterday and I've been thinking about it all day because this launch tells you everything about where AI coding tools are heading — and honestly, it doesnt look great for Cursor specifically even though the product itself is kind of impressive.
Heres what happened. Cursor killed the code editor. Not literally, its still there, but the default view in Cursor 3 is now an agent orchestration panel. No file explorer front and center. No code-first layout. You type what you want in natural language, hit enter, and AI agents go build it. You can spin up multiple agents at once, watch them work in a sidebar, and review what they did. The whole thing was built under the internal codename "Glass" and its basically Cursor admitting that the product that made them famous — the AI-powered code editor — isnt the future anymore. Jonas Nelle, one of their heads of engineering, told WIRED straight up: "A lot of the product that got Cursor here is not as important going forward anymore." Thats a wild thing to say about a product thats generating $2 billion in annualized revenue.
The reason they did this is obvious if youve been paying attention. Claude Code owns 54% of the AI coding market now according to Menlo Ventures data. Claude Code and OpenAI's Codex both let you spin up agents that work for hours without supervision, and theyre both offered through $200/month subscriptions that give you way more than $200 worth of compute. WIRED reported that Claude Code and Codex users regularly get over $1,000 worth of usage on those plans. Anthropic and OpenAI can afford to burn cash on customer acquisition because theyve raised hundreds of billions between them. Cursor raised $3 billion total, which sounds like a lot until you realize Anthropic alone is valued at $380 billion.
And developers are noticing. Multiple people told WIRED theyve shifted most of their coding work to Claude Code and away from Cursor. One founder said his decision basically comes down to whoever has the most generous rate limits. Another said he rarely touches Cursor anymore despite using it heavily last year. The thing that made Cursor special — being the best IDE with AI built in — stopped mattering when the AI got good enough to just build the whole thing without an IDE.
But heres where it gets messy. Cursor launched Composer 2 to power all this, and they claimed it matches GPT-5.4 on coding benchmarks at one-tenth the inference cost. Sounds amazing right? Except TechCrunch reported that Composer 2 is actually built on top of Moonshot AI's open-source Kimi 2.5 model, and Cursor didnt disclose that until users pushed them on it. Thats not illegal or anything — Kimi 2.5 is open source — but not saying "hey this is fine-tuned Kimi" upfront when youre marketing it as your own model is the kind of thing that makes developers trust you less. And trust is basically all a dev tools company has.
The pricing situation is also kind of insane. One early reviewer burned through roughly $2,000 in two days of normal use with Cursor 3. Two thousand dollars. In two days. Meanwhile that same workload runs at a flat $200/month on Claude Code with unlimited Opus access. Cursor tried usage-based pricing back in mid-2025 and developers hated it so much the company had to apologize. Now theyre doing it again but with higher stakes because the alternative products are better and cheaper.
I use Claude Code for most of my stuff tbh and the reason is exactly what everyone else is saying — the value per dollar is absurd right now. I can spin up background agents that work on different parts of a project simultaneously while I review what the previous batch produced. Cursor was amazing when AI coding meant autocomplete and inline suggestions. But we're in what Cursor's own CEO Michael Truell calls the "third era" — first was autocomplete through 2025, second was synchronous copilots where you guided the AI, and now its autonomous agents that work independently for hours. The problem for Cursor is that the companies who make the actual AI models are naturally better positioned for that third era than a company that wraps those models in a nice interface.
That said, Cursor isnt dead and anyone saying that is prob being dramatic. They have $2 billion in revenue, 67% of the Fortune 500 as customers, and theyre generating 150 million lines of enterprise code per day. Their internal engineering team already has 35% of pull requests generated by autonomous agents running on cloud VMs — each agent gets a full dev environment, tests its output by navigating the UI like a human, and returns merge-ready code with video demos attached. That's genuinely cool and it shows the product works when it works.
The real question is whether a $29 billion coding startup can survive when Anthropic and OpenAI are willing to subsidize their competing products indefinitely. Cursor's head of engineering recently left. Fortune reported that several startups in one investor's portfolio are activley moving off the platform. The company is trying to raise at a $50 billion valuation right now, which either means they're confident or desperate, and in this market its honestly hard to tell which.
My take: Cursor 3 is a good product launched from a position of weakness. They built exactly what they needed to build — an agent-first interface that competes directly with Claude Code and Codex — but theyre doing it 6 months late with a model wrapped around someone else's open source project and pricing that makes developers do math before pressing enter. The AI coding war is real, its happening right now, and the companies with the deepest pockets and the best models are winning. Cursor's still in the fight but the clock is ticking and everyone including them knows it.
Top comments (0)