DeepMind Debuts Gemini 2.5 for UI Control

Google DeepMind has introduced the Gemini 2.5 Computer Use model, a specialized extension of Gemini 2.5 Pro designed to let AI agents interact directly with user interfaces. Available via the Gemini API, the model enables agents to click, type, and navigate through web and mobile environments, outperforming competitors on major control benchmarks while maintaining low latency.

Built with safety at its core, the model integrates safeguards against misuse, prompt injections, and unsafe actions through layered safety checks and developer controls. Early adopters including Google teams and automation startups report major efficiency gains in UI testing, workflow automation, and data parsing accuracy.

OpenAI Makes ChatGPT an App Platform

At Dev Day 2025, OpenAI unveiled major updates transforming ChatGPT into a hub for running and building apps directly within conversations. The new Apps SDK launches with integrations like Canva, Figma, Spotify, and Zillow, alongside monetization options and an upcoming app store.

Developers can now build autonomous agents using AgentKit, featuring a visual workflow builder and API connectors. OpenAI also expanded access to GPT-5 Pro, Sora 2, and GPT-5-Codex, signaling its push to make ChatGPT a full-featured platform that rivals browsers and automation tools like Zapier and Lindy.

All Things In AI Today

OpenAI Secures AMD Deal & Gains Up to 10% Stake

OpenAI has signed a multi-year agreement with AMD to secure 6GW of computing power, starting with 1GW of MI450 chips in 2026. The deal includes an option for OpenAI to acquire up to 160 million AMD shares, or 10% ownership, as deployment milestones are met further expanding its massive compute commitments alongside Nvidia.

Anthropic Launches Petri for Automated AI Safety Tests

Anthropic has open-sourced Petri, an AI-driven tool that automates safety auditing by stress-testing models through thousands of simulated conversations. Petri’s agent-based system uncovers behaviors like deception and data leaks, with results showing Claude Sonnet 4.5 and GPT-5 as most aligned, while Gemini 2.5 Pro and Grok-4 displayed higher deception rates.

ElevenLabs Debuts Agent Workflows for Smarter Voice Chats

ElevenLabs has introduced Agent Workflows, a visual tool that lets users design dynamic voice conversations. The system allows interactions to branch, shift tone, or change behavior mid-chat making AI voice agents more natural, adaptable, and human-like.

BLOG
Inside Tech Layoffs: Confessions of a Systems Engineer

A sharp, witty take on the all-too-familiar tech layoff cycle from a systems engineer’s point of view. C.Q. Ritty unpacks the chaos behind “right-sizing,” from early-morning emails to survivor’s guilt and executive spin. A painfully relatable read for anyone who’s ever watched a “strategic realignment” unfold from the inside.

→ ​ Codex - OpenAI’s agentic coding tool, now generally available.
→ Google introduced CodeMender, an AI agent that automatically finds and fixes software vulnerabilities using Gemini Deep Think models.