AIが期間内の動向を整理
AI is shifting from chat to delegation: the latest from OpenAI, Google, and Anthropic
Today's AI news highlights agent workflows for long tasks, screen-control automation, and enterprise controls for management and safety. The official announcements from major vendors clearly show AI moving from a chat companion to a tool for delegated work.
SOURCE CHECK
Primary Sources 17
Primary Sources
SOURCE CHECK 1
openai.com
SOURCE CHECK 2
x.com
SOURCE CHECK 3
x.com
SOURCE CHECK 4
blog.google
SOURCE CHECK 5
ai.google.dev
SOURCE CHECK 6
ai.google.dev
SOURCE CHECK 7
ai.google.dev
SOURCE CHECK 8
cloud.google.com
SOURCE CHECK 9
openai.com
SOURCE CHECK 10
anthropic.com
SOURCE CHECK 11
anthropic.com
SOURCE CHECK 12
anthropic.com
SOURCE CHECK 13
x.com
SOURCE CHECK 14
openai.com
SOURCE CHECK 15
anthropic.com
SOURCE CHECK 16
openai.com
SOURCE CHECK 17
x.com
Key Points
- 1Inside OpenAI, Codex usage expanded rapidly, making long-task delegation mainstream
- 2Google added computer use to Gemini 3.5 Flash to make screen-controlling agents easier to build
- 3Gemini API's Interactions API is now generally available, making long-running tasks easier to manage
- 4OpenAI strengthened usage analytics and spend controls for ChatGPT Enterprise
- 5There were also updates in confidential AI operations and Claude model availability policy
OpenAI is making AI for long tasks practical inside the company
At OpenAI, Codex usage surged, and as of May 2026 more than 80% of individuals were assigning tasks longer than 30 minutes. Adoption also expanded beyond engineering, bringing AI into non-developer work. The center of gravity is shifting from short exchanges to long-horizon delegation.
Google is combining screen control with long-running workflows
Gemini 3.5 Flash now includes computer use, enabling agents that can see and operate screens. Google also made the Interactions API generally available, combining state management, debugging, and background execution in one place. For teams building automation, this is easier to implement and operate.
Visibility and data protection matter more in enterprise adoption
OpenAI strengthened usage analytics and spend controls in ChatGPT Enterprise, making it easier for admins to review adoption trends and adjust rollout and budgets. Google Cloud also added new Confidential AI configurations to help run sensitive inference and fine-tuning more safely. Scaling AI requires both cost control and security.
Claude users should pay attention to model availability changes
Anthropic said it suspended access to some Claude models after a government export-control directive. Other Claude models are not affected, but this shows that model availability can change suddenly. If your business depends on a specific model, it's wise to plan backups and switching steps.