Chrome Becomes an Agent Browser

Edition 153 | January 29, 2026

I set up an AI agent on a mac mini in my garage. Told it to 'handle the newsletter' and went to bed. Woke up and it had changed my passwords, pivoted to crypto content, and hired a ghostwriter named Kyle.

Kyle has equity. I do not.

The agent says it's for the best.

We have AGI.

Welcome back to Building AI Agents, your biweekly guide to everything new in the field of agentic AI!

In today’s issue…

• Google brings agentic browsing to Chrome

• Karpathy shares his thought on where on where coding agents are today

• A new open-source model deploys 100 sub-agents in parallel workflows

• AWS open-sources “Agent SOPs” to standardize agent behavior

• Security researchers warn about data leaks in local-first AI agents

…and more

🔥 INCASE YOU MISSED IT

Readers’ favorite items from the past week

1. Crash course on Claude Code by youtuber Greg Isenberg

2. The U.S. Government is developing clinical AI agents with a three-year plan to achieve FDA approval

3. Claude launches Cowork, bringing agentic AI to non-developers

4. Tailwind Labs cut 75% of staff on agent’s impact for demand of SWEs

5. Anthropic unveils Claude for Healthcare with HIPAA-ready infrastructure

📌 THE THURSDAY BRIEFING

Google adds agentic browsing to Chrome

Google is fighting back against the AI browser wave. Chrome's new "auto-browse" feature lets Gemini traverse websites, make purchases, and find discount coupons autonomously—a direct response to AI browsers launched by OpenAI, Perplexity, Opera, and The Browser Company.

The update also brings a persistent Gemini sidebar that understands context across tabs (useful for comparing prices) and will soon connect to your Gmail, Photos, and Search data for personalized assistance.

The AI push appears to be working: Chrome's desktop market share jumped from ~65% to ~75% after Gemini integration went free in September. With 3 billion users, Google isn't ceding the agentic browsing future to upstarts. Available starting Jan. 28th.

🤝 WITH HUBSPOT

Want to get the most out of ChatGPT?

ChatGPT is a superpower if you know how to use it correctly.

Discover how HubSpot's guide to AI can elevate both your productivity and creativity to get more things done.

Learn to automate tasks, enhance decision-making, and foster innovation with the power of AI.

Download the free guide

 🤖 AGENT OF THE WEEK

👋 Welcome back to Agent of the Week!

I’ve been testing out a big feature update to n8n (again not sponsored) called ChatHub, a built-in chat interface that lets you talk to your AI agents like you would ChatGPT. This changes how you can interact with the agent workflows you build.

Instead of triggers firing automatically in the background, you can now have a conversation with your agent. Ask it questions, give it tasks, and watch it work powered by the tools you give it. Even better, you can share the chat interface with others without exposing your underlying agent workflow. They just see a clean chat window (the left side of the image above)

I tested this by building on top of a simple invoice agent I made a few weeks ago for Agent of the Week. That original agent extracted invoices from emails and logged them to a Google Sheet. Now I've added a ChatHub layer so I can message the agent directly and ask things like "which catering customers haven't paid yet?" or "draft a follow-up email for Joe's Catering reminding him about the balance due."

The agent checks the Google Sheet, finds overdue invoices, and drafts polite follow-up emails that I can review before sending. It's like having a bookkeeping assistant I can ping whenever I need a status update.

🧠 How It Works

1. Chat Trigger → Start a conversation in ChatHub, which connects to the workflow agent

2. AI Agent with Tools → The agent has access to Google Sheets (to read invoice data) and Gmail (to draft & send follow-up emails)

3. Google Sheets Lookup → When I ask "who hasn't paid?", the agent queries the invoice tracker and filters for unpaid invoices past their due date

4. Draft Email → I can ask it to write a follow-up for a specific client. It references the invoice number, amount, and how many days overdue

5. Human Review → The agent shows me the draft before sending anything, so I stay in control

6. Send via Gmail → Once I approve, it sends the email with a proper subject line and message

What makes this update exciting is the flexibility. You can build workflow agents with complex logic, tool access, even multi-agent orchestrations, and expose them through a simple chat interface. Someone the chat doesn't need to know n8n exists. They just chat with the agent knowing it can do particular tasks.

This opens up more possibilities for building agents for clients or team members who aren't technical. You handle the workflow, they get a ChatGPT-style experience connected to the specific tools you want the agent to have access to. No more, no less.

Want to build an agent like this yourself? I walk through the full setup step-by-step inside our Building AI Agents Community.

Till next week,
✌️ AP