Here’s your AI news this week: three big announcements landed within days of each other, and they point at the same thing: AI is getting cheaper, more autonomous, and more hands-on with your actual work. Here’s what happened.
Google puts an agent inside your company data
At Gemini at Work 2026, Google Cloud CEO Thomas Kurian introduced the Gemini Agent. It’s an autonomous agent built into Google Cloud and Google Workspace. It connects to corporate data stores and to outside platforms like SAP, Salesforce, and Workday, then carries out multi-step tasks across them.
The detail I’d pay attention to is that it queries databases where they sit. It doesn’t copy or move sensitive records to do its work. If you’ve ever sat through a security review for an AI tool, you know that’s the question that usually kills the project.
This is Google going straight at Microsoft Copilot Autopilot and OpenAI’s Agents API. Enterprise workflow automation is now a three-way fight, and nobody is pretending otherwise.
Sources: Google Cloud blog, Google blog, Google Cloud Press Corner
Claude Haiku 5.5 costs $0.10 per million input tokens
Anthropic released Claude Haiku 5.5 at $0.10 per million input tokens and $0.50 per million output tokens. Those are small numbers. The part that makes people look twice is the performance: it reportedly matches GPT-6 Luna and older flagship models on coding and reasoning benchmarks.
Anthropic built it for fast agent loops and large-scale data processing. That’s where cost hurts most. An agent that makes dozens of model calls to finish one task gets expensive quickly, and a background pipeline chewing through millions of documents does too. At these prices, a lot of “too costly to run” ideas start to look reasonable.
I’d still run it on my own workload before believing any benchmark table. But the pricing alone is going to push other labs.
Sources: Anthropic announcement, Simon Willison’s breakdown, Artificial Analysis
ChatGPT now builds small apps in the chat window
OpenAI is rolling out GPT-6 along with a feature called Intelligent UI, and it’s reaching everyone, free users included. Instead of answering only in text, ChatGPT can render working interface pieces inside the conversation: live calculators, interactive charts, clickable mini web apps.
Ask a question that needs a loan calculator, and you get a loan calculator. You don’t get a paragraph explaining how to make one. With over 1.2 billion users, that’s a lot of people who may never need to open a spreadsheet or hunt for a niche tool again.
Whether the generated apps hold up for anything beyond quick one-offs is an open question. Still, moving a chatbot from “writes text” to “makes software on the spot” is a real change in what the product is.
Sources: OpenAI release, OpenAI Developer Community, NDTV Profit
What ties these together
Google wants agents running inside your business systems. Anthropic wants them cheap enough to run all day. OpenAI wants the chat window to produce working tools. Put together, you get agents that act on your data, cost almost nothing per step, and hand you software when you ask for it. I’m curious which of the three will matter most a year from now, and my money is on the cheap one.