Cursor Announces Its Own AI Model, Git Platform, and Mobile App

Cursor, the AI coding tool, has announced its own AI model, a Git platform, and a mobile app all at once. The company had previously depended on external models from OpenAI, Anthropic, and others, and its decision to finally commit to an in-house model could reshape the competitive structure of the coding AI market.

The new Git platform competes with GitHub on code hosting, but it is notable for appearing to have been designed from the ground up for AI-native development workflows. Meanwhile, the mobile app means that AI pair programming, previously confined to the desktop, can now happen on the go.

The coding agent market has seen a flurry of moves this week, including the GA (general availability) of the GitHub Copilot App and the end of JetBrains' Junie beta. Cursor's strategy appears aimed at shifting from a mere editor extension to an entire development platform.

ByteDance's Seedance 2.5 Breaks the 30-Second Barrier in AI Video Generation

Seedance 2.5, ByteDance's video generation AI, has succeeded in generating continuous videos longer than 30 seconds. Most existing AI video generation tools have been limited to short clips of about 5 to 10 seconds, so 30 seconds is a length that can be directly applied to production workflows for advertising, social media content, and short promotional videos.

This result shows that technology for generating video from text or images is beginning to step into the territory of practical long-form content. Whether the quality of the generated videos is consistently stable and can withstand real-world production, however, remains to be verified.

JetBrains' AI Coding Agent Junie Exits Beta

Junie, the AI coding agent developed by JetBrains, has been released as a GA (general availability) version. JetBrains is a development tools vendor known for IntelliJ IDEA, PyCharm, and other IDEs, and its AI agent integration stands out for building on its existing IDE user base.

Junie is designed so that code generation, refactoring, test writing, and more can all be completed within the IDE. In contrast to the GitHub Copilot App, which ships as a standalone desktop application, Junie takes an approach of weaving AI into the existing IDE experience.

IKEA Turned 8,500 Jobs Displaced by AI into a €1.3 Billion Business

IKEA's case is drawing attention. With the adoption of AI, 8,500 of the company's employees moved away from their previous duties, but instead of layoffs, they were reassigned to a new business division. As a result, a new business worth €1.3 billion (approximately 210 billion yen) has reportedly been created.

The case shows that AI-driven job displacement does not necessarily mean headcount cuts. That said, IKEA was able to pull this off thanks to large-scale retraining investment and the specific industry structure of furniture and retail — it is worth noting that not every company can replicate the same approach.

At the same time, skeptical voices about AI productivity gains persist. On Hacker News, a lively discussion asking "If AI supposedly makes us 10x more productive, why are coding agents still awkward to use?" points to a gap that still remains between tool maturity and expectations.

Police Officer Under Criminal Investigation Over AI-Generated Evidence — The Societal Risks of AI

A police officer in the UK is under criminal investigation on suspicion of using AI-generated evidence in legal proceedings. The case has raised serious concerns as a possible instance of AI-generated content being improperly used in real-world legal and administrative processes.

The case underscores the reliability of AI-generated material and the importance of mechanisms for verifying it. As deepfakes and AI-generated text make it easy to produce content that merely looks plausible, how to guarantee the authenticity of information remains an unsolved problem, both technically and institutionally.

IBM Research's CUGA: 24 Practical Agent App Examples

IBM Research has published 24 practical agent app examples built with CUGA, a lightweight agent harness. Rather than abstract debate, the collection demonstrates agent design patterns through working code examples.

Designing agent apps involves many challenges that surface in real-world operation, such as tool selection accuracy, error recovery, and context management. The CUGA examples serve as a useful reference for concrete solutions to these challenges. They are published on the Hugging Face Blog and can be explored along with the code.

---