OpenAI's Codex Gains Windows Control, Salesforce Reports Gains with Claude Code

OpenAI's Codex can now autonomously operate Windows PCs for testing, while Salesforce reports significant productivity gains using Anthropic's Claude Code for development migrations.

The AI landscape continues to evolve with advancements in autonomous operation and reported productivity gains. OpenAI's Codex is now capable of independently controlling Windows PCs, opening new avenues for automated testing and bug hunting. Meanwhile, Salesforce reports substantial improvements in development efficiency after migrating its dev org to Anthropic's Claude Code. These developments highlight the increasing integration of AI into critical workflows.

💻 OpenAI's Codex Now Operates Windows PCs Autonomously

OpenAI's Codex application can now independently control Windows 11 programs, test applications, and hunt for bugs, according to *The Decoder*. This new "Computer Use" feature allows the AI to operate autonomously, even when no one is physically at the PC. Users can remotely start and monitor tasks from their phone using the ChatGPT mobile app. This capability opens up new possibilities for automated software testing and bug detection, potentially streamlining development workflows.

🚀 Salesforce Reports Productivity Gains with Claude Code Migration

Salesforce reports significant productivity gains after migrating its entire dev org to Anthropic's Claude Code, as reported by *The Decoder*. The company claims a 79 percent increase in pull requests per developer and a five percent reduction in incidents for April 2026. The migration, which involved removing token limits, reportedly reduced a 231-day migration process to just 13 days. While these numbers are not independently verifiable, they suggest a substantial impact on development efficiency.

✨ OpenAI Upgrades GPT-5.5 Instant, Retires Older Models

OpenAI is enhancing GPT-5.5 Instant with improved readability and phasing out older models, according to *The Decoder*. The update aims to provide more natural responses, while the Canvas feature is being removed, with writing and coding tasks now running directly in the chat interface. Additionally, the older o3 and GPT-4.5 models from ChatGPT will be retired by August 2026. This move reflects OpenAI's ongoing efforts to refine its models and focus on the latest generation of AI technology.

🤔 AI Search Agents Rely Heavily on Existing Knowledge

Leading AI search agents like GPT-5.4 and Kimi K2.6 often rely on pre-existing knowledge rather than conducting genuine web research, according to *The Decoder*. Researchers at the Harbin Institute of Technology found that these models primarily use the web to confirm information they already learned during training. When tested with a new time-based benchmark called LiveBrowseComp, which focuses on events from the last 90 days, the models' performance significantly declined, suggesting a limited ability to research novel information.

📐 Terence Tao Envisions Division of Labor in Mathematics with AI

Mathematician Terence Tao suggests that AI could revolutionize math research by enabling a division of labor, a concept previously absent in the field, according to *The Decoder*. Traditionally, researchers have had to master every step of the process themselves. Tao envisions the emergence of "industrial mathematics," where large, AI-supported teams collaborate, with humans remaining crucial for generating "inspired guesses." This shift could potentially accelerate mathematical discovery and innovation.

What this means: The advancements in OpenAI's Codex and the reported gains at Salesforce demonstrate the increasing practical applications of AI in software development and enterprise operations. The ability of AI to autonomously operate computers and accelerate development processes signifies a shift towards greater automation and efficiency. However, the limitations of AI search agents in conducting original research highlight the ongoing need for human oversight and critical evaluation.

The industry continues to move towards more integrated and autonomous AI solutions, with a focus on enhancing productivity and streamlining complex tasks.