Codex Reaches 7 Million Users — Competition with Claude Code Heats Up
According to a Latent Space report, OpenAI's Codex has grown more than 10x over the past six months, reaching 7 million users. A reported 1 million new users were added in a single day.
Codex's rapid growth signals that the AI coding tool market is expanding fast. Anthropic, meanwhile, has not officially disclosed Claude Code's user numbers, so a direct share comparison is difficult at this point. Latent Space says it will "examine quiet-day data to explore whether Codex has overtaken Claude Code."
In AI coding, choosing a tool is no longer just a matter of preference — it's becoming a decision that affects the efficiency of the entire development process.
RCE Vulnerability in Meta's Manus AI — The Permission Risk of AI Agents
An article published on Zenn revealed that a remote code execution (RCE) vulnerability existed in Meta's Manus AI desktop app. The finder was awarded a $4,500 bounty.
Making AI agents practical requires granting them certain permissions, but those permissions can be abused through attacks such as prompt injection. In the Manus AI case, packaging the agent as a desktop app gave it access to the local system, expanding the attack surface.
Permission design for AI agents presents the fundamental trade-off between functionality and security, and implementing sandboxed execution and permission separation has become urgent.
Rating the "AI Readiness" of 31 Japanese SaaS Companies
An article on Zenn examined the AI readiness of 31 major Japanese SaaS companies and published ratings for 26 of them, graded from AAA to D. On July 1, 2026, Gartner predicted that enterprise software is entering the era of "buying for your agents," with roughly 20% of SaaS spending ($234 billion) affected by 2030.
According to the findings, freee and Money Forward in the accounting space, along with the Cybozu family of products, are leading the way. Meanwhile, many companies in HR, contracts, and payments remain at the stage of "the API exists, but there's no AI-facing entry point (official MCP) yet." Only a small fraction of the 31 companies offer an official MCP.
For SaaS to be "chosen by AI," building interfaces for AI agents — not just human-facing UIs — is becoming essential.
What Google's Gemini 3.5 Flash Reveals About Search Becoming an AI Assistant
Google has announced a new AI model, Gemini 3.5 Flash. The improved model performance is drawing attention, but the more significant change is that the act of "searching" itself is starting to transform.
Traditionally, search was a process in which users came up with keywords themselves and hunted for information while comparing multiple websites. The Gemini 3.5 Flash update suggests an evolution toward AI organizing information on the user's behalf and presenting context-aware answers.
This shift, often called the "AI assistantification of search," could transform how we access information altogether.
Why US Open-Source AI Lags Behind China
A lively debate is underway in Reddit's LocalLLaMA community asking why US open-source AI labs are falling behind their Chinese counterparts on benchmarks.
While China's DeepSeek, Qwen, Tencent, and others dominate the top benchmarks with open-source models, US open-source labs have yet to release models that rival them. Cited factors include the gap in compute resources, differing corporate stances on open-source strategy, and structural issues in the research community.
The US-China race for open-source AI leadership isn't just a matter of technical capability — it also reflects differing philosophies in how the entire ecosystem is designed.
Zenveil: Security Scanning for AI-Generated Code
Hacker News featured Zenveil, a security scanning tool for AI-generated code. As AI coding tools spread, demand is growing for automatically detecting vulnerabilities lurking in generated code.
Zenveil scans AI-generated code for known vulnerability patterns and improper implementations, aiming to build security checks into the development process.
How to guarantee the quality and safety of AI-generated code remains one of the key challenges on the road to serious adoption of AI coding tools.
---