Cursor Launches a New AI Agent Experience to Take On Claude Code and Codex https:// fed.brid.
agents anthropic claude cursor openai sora
| Source: Mastodon | Original article
Cursor unveiled a new version of its development platform, Cursor 3, on Thursday, adding a suite of “always‑on” AI agents that can write, review and refactor code without a manual prompt. The flagship model, dubbed Composer 2, is positioned as a low‑cost alternative to Anthropic’s Claude Code and OpenAI’s Codex, promising comparable accuracy while running continuously in the background to handle tasks such as pull‑request reviews, bug triage and incident response.
The rollout marks Cursor’s shift from a single‑assistant workflow to a multi‑agent architecture that can be instantiated on demand through a simple UI toggle. Agents are pre‑trained on a curated corpus of open‑source repositories and can be chained together, allowing developers to hand off a high‑level goal—e.g., “implement OAuth login” —and let the system orchestrate code generation, linting and test creation autonomously. Pricing is tiered at $0.03 per 1 k tokens for the base model, with a premium “Pro” tier offering higher throughput for enterprise pipelines.
Why it matters is twofold. First, the move intensifies the race for the most accessible, production‑ready coding AI, a market that has seen rapid adoption of Claude Code despite the usage caps we reported on 2 April, when developers began hitting limits faster than anticipated. Second, Cursor’s always‑on agents blur the line between assistant and autonomous worker, raising questions about code provenance, security and the need for robust oversight mechanisms—issues already surfacing in safety‑aware multi‑agent frameworks for health communication.
What to watch next includes performance benchmarks against Claude Code and Codex, especially in large‑scale codebases, and how quickly development teams adopt the autonomous agents. Cursor has hinted at tighter IDE integrations and a forthcoming “audit‑log” feature to track agent actions, which could become a de‑facto standard for responsible AI‑driven development. The next few weeks will reveal whether the low‑cost proposition translates into measurable productivity gains or merely adds another option to an increasingly crowded field.
Sources
Back to AIPULSEN