NewApp Digest
search bot

AI coding

Google’s Android Bench 2.0 tests AI agents on week-long app builds
AppWizard
September 19, 2026

Google’s Android Bench 2.0 tests AI agents on week-long app builds

Google has launched Android Bench 2.0, an upgraded benchmark for evaluating AI coding agents, which now includes 30 long-horizon tasks that can take human engineers days or weeks to complete. The best pass rate for these tasks is around 28%, down from approximately 91% in the previous version. The benchmark measures both pass rates and completion rates, acknowledging partial progress rather than just failures. It assesses various challenges, including app creation and feature integration, using a comprehensive scoring methodology that evaluates functionality, regression checks, and visual fidelity. AI models perform better with new code than with modifications to existing systems, facing challenges in runtime validation and cross-platform conversions. The current leaderboard shows OpenAI’s GPT-6 Astra leading with a 28% pass rate, while Gemini 3.8 Flash has an 8% pass rate. Developers using AI agents should be aware that while AI can generate significant portions of applications, further refinements will require human input. The benchmark is available on Google’s Android Bench leaderboard.
⚡ Weekly Recap: VMware Exploits, Windows 0-Day, MCP Attacks, Browser Hijacks and More
Winsage
August 17, 2026

⚡ Weekly Recap: VMware Exploits, Windows 0-Day, MCP Attacks, Browser Hijacks and More

A suspected advanced persistent threat (APT) group linked to China exploited a newly patched vulnerability in VMware vCenter (CVE-2026-59310), which has a critical CVSS score of 9.8, allowing for arbitrary code execution and the deployment of Babuk-derived ransomware. A recently patched vulnerability in Apple macOS (CVE-2026-65400) has been exploited to deploy a cryptocurrency miner, granting unauthorized root access. The Lazarus Group from North Korea exploited a zero-day vulnerability in Microsoft Windows, targeting defense and aerospace sectors. GeoServer patched a critical SQL injection vulnerability that was actively exploited. A new macOS malware, Amnesia Stealer, targets users through ClickFix attacks, stealing data and allowing real-time access to authenticated sessions. A novel attack technique named GhostSplice can manipulate AI coding assistants. Research revealed a method exploiting Chromium's DevTools Protocol for data theft. Noteworthy CVEs this week include CVE-2026-68820, CVE-2026-58231, and multiple others across various platforms. A high-severity command injection flaw in FileRun allows remote code execution. An advanced ClickFix attack has been reported, deploying sophisticated malware. A heap overflow vulnerability in Citrix NetScaler was patched after indications of exploitation. A new malware loader targeting Portuguese-speaking users has been identified. A significant reduction in exposed Automatic Tank Gauge systems has been observed. A phishing campaign targeting Brazil has been detected, and an F.B.I. agent faces charges for unauthorized crypto withdrawals. Authorities in Ukraine dismantled fraudulent call centers, and a North Carolina man was sentenced for cyber extortion. Unauthorized access to sensitive data by the ExfilSquad group has been confirmed. LightSpy activity linked to China has been detected in over 13 countries. A supply chain attack exposed over 2,500 companies, and an Azure exfiltration campaign has exposed millions of enterprise records.
After rewriting SQLite in Rust, Turso turns its sights on Postgres
Tech Optimizer
July 29, 2026

After rewriting SQLite in Rust, Turso turns its sights on Postgres

Cloud database provider Turso is developing a Postgres-compatible implementation based on its SQLite-compatible database, which was created from scratch in Rust. CEO Glauber Costa believes Postgres can benefit from modernization for cloud-native applications. Turso's SQLite reimplementation, now called Turso, uses a virtual machine architecture that may eventually support other database frontends, including MySQL and Redis. The company initially forked SQLite into libSQL but later pivoted to a cloud service named Turso. They focused on a virtual machine architecture for their rewrite, which translates SQL queries into a custom bytecode language. Turso has developed a Postgres-compatible prototype named pgmicro, which aims to run existing applications with minimal modifications. The company is also working on a database-as-a-cloud service that will allow customers to use various database types on the Turso platform.
Unity demonstrates 'an Unreal character running through a Unity world' for the first time
AppWizard
July 22, 2026

Unity demonstrates ‘an Unreal character running through a Unity world’ for the first time

Unity has demonstrated a Unity Engine game operating within an Unreal Engine environment at the Unite Seoul keynote. This collaboration with Epic Games aims to allow Unity developers to publish games directly within Fortnite. Epic Games CEO Tim Sweeney expressed support for this integration, highlighting the importance of connectivity among game engines. Unity's Fortnite integration is set to enter early access in 2027, and Unity 7, the next version of its game engine, will be released next year, featuring enhanced rendering, improved monetization options, collaboration tools, and a protocol for integrating generative AI coding agents.
The TechBeat: How Small Postgres Metadata Tables Quietly Throttle Your Largest Queries (7/6/2026)
Tech Optimizer
July 7, 2026

The TechBeat: How Small Postgres Metadata Tables Quietly Throttle Your Largest Queries (7/6/2026)

- Websites detect VPN usage through factors like IP reputation, DNS leaks, WebRTC, and browser fingerprints, with strategies to mitigate tracking. - Slow Postgres dashboards may be caused by stale metadata statistics rather than the fact table, with solutions provided to rectify these issues. - Spec-driven development enhances the reliability of AI coding agents through structured specifications, workflows, reusable skills, and proof. - A comparison of usage-based billing platforms for 2026 includes Credyt, Metronome, Orb, Lago, Stripe Billing, and Chargebee. - Loop Engineering is a rising trend in AI workflows expected to dominate in 2026, though challenges may accompany it. - ONLYOFFICE released version 3.7 of DocSpace Developer, introducing AI file generation, new webhook events, expanded plugin tools, and enhanced form workflows. - ANY.RUN's data inspection tool offers SOC teams comprehensive browser visibility for quicker phishing investigations. - AI systems are reducing obvious hallucinations, but deeper structural failures remain a risk for professionals. - A vendor audit revealed supply chain vulnerabilities in AI, including unpinned models and the use of trust_remote_code=True. - Developers using OpenAI Codex should avoid overloading a single conversation with tasks to maintain performance, recommending the use of subagents. - Barriers to AI adoption are not purely technological, as shared by insights from discussions with 500 individuals. - Guidance on creating production AI agent loops emphasizes state management, tool governance, evidence verification, and domain-specific controls. - Dawnguard launched a security architecture automation platform with million in pre-seed funding to address AI code vulnerabilities. - The AI boom is driven by complex memory architecture rather than GPUs, influencing consumer electronics pricing strategies. - The origins of functional programming trace back to John McCarthy's practical needs for differentiating expressions. - A guide for AI startup founders addresses common pitfalls in understanding AI agents and the autonomy trap for developing reliable systems.
Microsoft is rewriting Windows 11 shell in native code, killing the web slop slowing your PC
Winsage
June 4, 2026

Microsoft is rewriting Windows 11 shell in native code, killing the web slop slowing your PC

Microsoft is enhancing Windows 11 by reducing reliance on web-based components, which have caused performance issues. The user interface framework WinUI will now be referred to simply as "WinUI," indicating no plans for a WinUI 4, to reassure developers. Microsoft is integrating WinUI into the Windows 11 shell more rapidly, moving away from web components. A specialized team is focused on creating "100% native" Windows applications, starting with an overhaul of the Start menu from React Native to WinUI components. Performance, memory optimization, and quality control are top priorities for the WinUI team. A new system compositor is being introduced to improve UI performance, and enterprise-focused DataGrid and Charting controls are being added to WinUI. Microsoft is expanding open-source development for WinUI, allowing public testing and contributions. Migration support between WinForms and WinUI is being enhanced for legacy applications. An experimental project called Microsoft UI Reactor is introduced to simplify native application development using C#. This shift towards native development aims to improve performance and efficiency for all Windows users.
Google is reportedly buying Play Store app code from devs, and you can guess why
AppWizard
June 3, 2026

Google is reportedly buying Play Store app code from devs, and you can guess why

Google has initiated a confidential program to acquire app source code from select developers on the Play Store to enhance its AI-powered coding tools. The company is offering financial incentives to developers who share their app source code, allowing them to retain ownership and intellectual property rights. Google is specifically targeting real-world Android applications, including older projects, to improve its Gemini models and coding functionalities. The initiative aims to elevate the performance of Google's AI offerings, particularly the upcoming Gemini 3.5 and Antigravity 2.0 coding agent, and to position the company more competitively against rivals like Microsoft and Anthropic.
Search