Google upgrades Opal with AI agents

Cursor wants agents to act like developers

Cursor says 35% of its pull requests now come from AI agents running on their own virtual machines.

The $29.3 billion startup rolled out a major update that lets its coding agents test their own changes, record their work through videos and logs, and run in parallel in the cloud.
They can be triggered from Slack, GitHub, mobile, or Cursor’s desktop app.

This is a shift from AI as autocomplete to AI as worker.
When an agent can spin up its own full development environment and iterate until a feature works, the bottleneck moves from typing code to deciding what should be built in the first place.

Competition is tight. Anthropic’s Claude Code is at a $2.5 billion run rate, OpenAI’s Codex has 1.5 million weekly users, and GitHub Copilot counts 26 million users.
On a quiet afternoon with laptops humming, the real contest is about throughput, and whether teams trust agents to operate without constant supervision.

If developers get comfortable reviewing instead of writing, team structure changes fast.

Opal moves from app builder to task runner

Google just gave Opal an agent that can plan and execute tasks on its own using Gemini 3 Flash.

The new feature lets users create mini apps that choose tools automatically, store memory in Google Sheets, and ask follow up questions when more input is needed.
Opal, launched in July 2025, is now available in over 160 countries and sits inside the Gemini web app.

This is Google trying to make “build an app” mean “describe a job.”
If the agent can plan next steps and manage state across sessions, the real value shifts from layout and buttons to how well the workflow actually runs.

The field is crowded. Startups like Lovable, Replit, Wabi, Emergent, and Rocket.new are chasing the same prompt to app idea. Late at night with a browser open, the difference will come down to which tool actually finishes the task without constant fixes.

If Opal handles execution reliably, it becomes more than a toy builder

The Pentagon wants full access to Claude

The Pentagon has given Anthropic until Friday evening to remove limits on how its AI models can be used across military applications.

In a fluorescent lit Pentagon conference room, Defense Secretary Pete Hegseth told CEO Dario Amodei the company could be pushed out of the supply chain.
Friday is the deadline.

Anthropic draws red lines around autonomous targeting decisions and mass domestic surveillance, while the Pentagon says it should be able to use any contracted model for all lawful national security purposes.
If the company refuses, officials say they could invoke the Defense Production Act and label it a supply chain risk, effectively forcing compliance on national security grounds.

This is escalating. Anthropic already holds a Pentagon contract worth up to $200m, alongside OpenAI, Google, and xAI. The real issue is control, and whether an AI company can meaningfully limit how the government deploys its tools once they are embedded in classified systems.

If the Defense Department wins this fight, other AI firms will be watching closely.

Prohuman team

Free Guides

Explore our free guides and products to get into AI and master it.

All of them are free to access and would stay free for you.

Feeling generous?

You know someone who loves breakthroughs as much as you do.

Share The Prohuman it’s how smart people stay one update ahead.