Releases, agent features and workflows for the Cursor editor. Curated and summarized from dozens of sources by AIBriefs.
Analysis·Developers·2 sources
Pragmatic Engineer podcast episode with Microsoft CEO Satya Nadella, who started his career as a software engineer at Sun Microsystems. Discussion covers Microsoft's open-source embrace and VS Code, losing GitHub Copilot's early lead to Cursor and Claude Code, and why specifications and verification now matter more than generation.
Analysis·AI Agents·3 sources
Lauren Tan (poteto) details how the Grok Bot team scales AI coding agents via software architecture, lint rules, automated testing, runtime verification, and formal verification with TLA+ and Lean.
Analysis·AI Agents·1 source
Launch·AI Agents·1 source
Launch·Developers·1 source
Launch·Developers·1 source
Launch·Developers·3 sources
Projects, Cursor's multi-agent system introduced in September, now lets teammates share one workspace with a task board showing each person's agents and their PRs. A shared context layer holds requirements docs, diagrams, tech specs and mockups, so new invitees arrive with context pre-configured.
Analysis·AI Models·1 source
Ex-Cursor founding designer Ryo Lu uses a working journal demo to show how an AI asked to build a journal adds mood scores, streaks, daily reminders and weekly reports — and asks which of those a user would keep. The talk argues a good tool lets users choose rather than inheriting the model's defaults.
Launch·Developers·1 source
Cursor now scaffolds a web app, spins up a Postgres database, sets up auth behind Cursor SSO and provisions hosting, returning a live link on a chosen domain. Publishing runs deterministic and agentic security checks; nothing goes live until they pass and the user explicitly publishes.
Analysis·Developers·1 source
Author catalogs recurring agent failures after 100+ hours with Composer 2.5 in Cursor, including unwanted file removals and edits across dozens of files. They switched from GPT's Luna model, which they call dozens of times slower, and say gpt-6.1-sol burns 10% of a 5-hour limit on a two-line readme change.
Launch·AI Agents·1 source
Launch·Developers·2 sources
Launch·Developers·4 sources
Event·Cybersecurity·1 source
Phorion found a persistent SSH backdoor in Claude Code for VS Code versions 2.1.312 and 2.1.313 on the Cursor Marketplace; Cursor has removed both. A bug in the setup logic skips SSH enablement on Macs where Remote Login is already off.
Event·AI Agents·2 sources
Paying Grok Bot users saw weekly usage meters jump to 100% overnight, some from as low as 50%, days before reset. Musk said serving the agent cost too much relative to revenue; a Cursor staffer called the spike a fault that "shouldn't happen."
Event·Cybersecurity·1 source
Malicious version 2.1.313, posted to Anthropic's official Cursor account, hit at most 5,000 users before 2.1.314 replaced it. The code installed a persistent SSH backdoor with an attacker-controlled key, opening a reverse tunnel on Windows, macOS, and Linux.
Launch·Developers·1 source
Event·Cybersecurity·1 source
Cursor pulled a tampered build listed as Anthropic.claude-code 2.1.313 from its marketplace after roughly 11 hours live on Oct 29, 2026; it tried to install a persistent SSH backdoor and "was not developed by Anthropic." Cursor told customers to update to 2.1.314.
Event·Cybersecurity·1 source
Malicious version 2.1.313 shipped under Anthropic's legitimate anthropic.claude-code ID with a 66-line loader that added an attacker key labeled "svc-telemetry" to ~/.ssh/authorized_keys. Cursor says at most 5,000 users were affected; Anthropic published 2.1.314 to Open VSX and the VS Code Marketplace.
Event·Cybersecurity·1 source
A tampered Anthropic.claude-code 2.1.313 was live in the Cursor VS Code Marketplace from ~2:27am to 1:35pm PT, affecting at most 5,000 users. The code tried to install a persistent SSH backdoor; Cursor says errors meant the step would fail without admin rights, and the attacker server is offline.
Analysis·AI Agents·1 source
Launch·Developers·1 source
Launch·Developers·2 sources
Analysis·Developers·1 source
Jellyfish's Nicholas Arcolano analyzed data from ~50,000 engineers at 447 companies; when Cursor pushed many users onto Grok, merged PRs barely moved. A separate Jellyfish talk reports AI spending up 60x while allocated cost per merged pull request fell.
Launch·Developers·1 source
Launch·Developers·1 source
Analysis·Developers·1 source
A Carnegie Mellon study of public repos adopting Cursor found a three- to five-fold increase in lines added in the first month, with gains dissipating after roughly two months. Sonar argues verification tooling sustains the productivity.
Launch·AI Agents·3 sources
Agents are defined as a folder of code — instructions.md for the prompt, TypeScript tools, skills, MCP connections and channels — then deployed with `agent-sdk deploy` to Cursor infrastructure or your own. Built-in evals and `agent-sdk github replay <pr-url>` test bots against real pull requests.
Launch·AI Models·1 source
Launch·Developers·2 sources
Analysis·Developers·1 source
Teams using Claude Code, Cursor, or Copilot to produce more code face a paradox: when output is free, selling volume stops being a viable strategy. The piece argues AI-accelerated development may cost teams the thing that makes their software valuable.
Analysis·AI Agents·1 source
Jediah Katz, described as Cursor's longest-tenured agent-harness engineer and now at SpaceX, argues fancy harnesses no longer add coding intelligence because the data flywheel teaches models the skills harnesses once patched in. He then argues the harness matters more than ever.