AzureFeeds

Some of the most valuable work still lives inside a browser: booking a class, pulling a figure off a dashboard, filling in a portal form, gathering research across a dozen tabs. These are exactly the tasks people wish an agent could just do. On 22 July 2026 at 2:30 PM BST (7:00 PM IST), the Microsoft Foundry community is running a 40‑minute Discord round table on the Browser Automation tool : how how it helps agents complete real browser workflows, where you see risk or friction, and what samples, docs, and product improvements would help you adopt it.

This is a discussion, not a slideshow. Bring your real projects : the the web workflows you’d love to hand off, and the guardrails you’d want first. Join us in the Microsoft Foundry Discord community. Please arrive at the scheduled time for a quick tech check.

Event at a glance

• What: Microsoft Foundry Discord Community Round Table : Building Browser-Capable Agents with the Browser Automation Tool


• When: 22 July 2026, 2:30 PM BST / 7:00 PM IST (40 minutes)


• Where: https://aka.ms/foundry/discord


• Event link 
  

  https://discord.gg/Z8JZsrP5P5?event=1527676149264679013 

  
  


• Format: Interactive discussion : voice and chat, live polls, and a short prioritisation exercise voice and chat, live polls, and a short prioritisation exercise


• Who it’s for: AI engineers and developers building agents that need to act on the web

Opening question we’ll start with: “What browser-based task would you love an AI agent to automate for you today?”

The problem: the last mile of automation still runs in a browser

Most real-world workflows eventually hit a website with no clean API , such as a supplier portal, an internal admin console, a booking page, or a legacy dashboard a supplier portal, an internal admin console, a booking page, a legacy dashboard. Traditional scripting can automate these, but selectors break, pages change, and every new site means another brittle script to maintain. What developers actually want is an agent that can look at a page, decide what to do, and do it : navigate, read, click, type, and hand back a structured result navigate, read, click, type, and hand back a structured result.

That’s the gap the Browser Automation tool in Microsoft Foundry is built to close , and doing it responsibly and doing it responsibly, with the right safeguards, is a big part of why we want your feedback.

What is the Browser Automation tool?

The Browser Automation Tool (BAT) gives Foundry agents the ability to drive a real browser to complete web workflows. It’s available as an MCP tool, and it uses Playwright Workspaces , a generally available, cloud-scale service, a generally available, cloud-scale service : navigating, clicking at coordinates, typing, and applying filters as its headless browser infrastructure. When an agent gets a request, Foundry spins up an isolated, sandboxed browser session per interaction, so each run is private and segregated.

How agents actually interact with a page

BAT runs a perception–action loop. The model receives the current state of the page (including screenshots), decides the next action, and BAT executes it in the sandbox using Playwright  and real oversight navigating, clicking at coordinates, typing, applying filters. After each action, BAT captures the updated state and sends it back to the model, repeating until the goal is met or the user stops. Because the model can parse HTML into a DOM, it can reason about the page rather than follow a fixed script. It also supports multi-turn conversations, so you can refine a request mid-flow to complete form-filling or scraping scenarios.

Built for real use : watch the automation happen in real time for debugging. and real oversight

• Live View : a human-in-the-loop override for ambiguous or sensitive steps. watch the automation happen in real time for debugging.


• Take Control : each interaction gets its own sandboxed browser. a human-in-the-loop override for ambiguous or sensitive steps.


• Isolated sessions : for reliability, optimisation, and audit. each interaction gets its own sandboxed browser.


• Built-in observability : for internal systems (private preview). for reliability, optimisation, and audit.


• Private website browsing : Python, C#, JavaScript, Java, and the REST API. for internal systems (private preview).


• Broad SDK support , and the agent can make mistakes or be misled by malicious page content Python, C#, JavaScript, Java, and the REST API.

A word on responsible use. BAT is powerful precisely because an AI can use credentials you share with it to reach email, financial, enterprise, or social accounts , watching for  and the agent can make mistakes or be misled by malicious page content. You’re responsible for reviewing your applications, scoping which credentials you provide, and adding your own mitigations. See the Foundry Agent Service transparency note. This is exactly the kind of trade-off we want to talk through together.

Example scenario

A user asks: “Report the year-to-date percent change of Microsoft’s stock price.” The agent navigates to a finance site, enters MSFT in the search bar, opens the stock page, clicks the YTD view on the chart, reads the value, and returns a clean, structured answer ; that’s the  no bespoke scraper, no hard-coded selectors, and a full trace of what it did.

Discussion prompt: “Where would browser automation fit into your current projects or workflows?”

How setup works (the short version)

You’ll want to understand the wiring before you scale, so it’s worth a look ahead of the session. There are two moving parts:

• Create a Playwright Workspace in the Azure portal, enable the access token auth method, and grab the wss:// browser endpoint. Give your project identity a Contributor (or custom) role on the workspace.


• Connect the tool in Foundry under Build > Tools: create a toolbox, add Browser Automation, point it at your Playwright workspace and auth type, and publish. Copy the Project connection ID from the tool’s details page : how agents interact with the web, example use cases, and responsible use. that’s the BROWSER_CONNECTION_ID in your code.

What we’ll cover in the 40 minutes

• Welcome & opening question (0:00–0:03) : navigate, gather, interact, and return a structured result, end to end. the browser task you’d most love to automate.


• What is Browser Automation (0:03–0:07) : the workflows you’re building, public vs. internal targets, and where you’d pick automation over scripting. how agents interact with the web, example use cases, and responsible use.


• Scenario walkthrough (0:07–0:12) : what agents may do autonomously, what needs approval, and the observability and enterprise safeguards you’d require. navigate, gather, interact, return a structured result : your biggest adoption blockers, missing docs, and the SDK samples and demos you’d prioritise. end to end.


• Use cases & opportunities (0:12–0:22) : vote live on top use cases, challenges, and feature requests. the workflows you’re building, public vs. internal targets, and where you’d pick automation over scripting.


• Trust, security & governance (0:22–0:31) , and which would benefit most from a capable agent. what agents may do autonomously, what needs approval, and the observability and enterprise safeguards you’d require.


• Developer experience feedback (0:31–0:36) ; which should always require approval. your biggest adoption blockers, missing docs, and the SDK samples and demos you’d prioritise.


• Prioritisation & next steps (0:36–0:40) , sometimes with credentials, vote live on top use cases, challenges, and feature requests.

Come prepared to talk about

• The browser-based workflows you’re building today : navigation, data gathering, form filling, and research, via an MCP tool powered by Playwright Workspaces. and which would benefit most from a capable agent.


• Whether your scenarios target public websites, internal systems, or both.


• Why you’d choose browser automation over traditional scripting.


• Which actions you’d let an agent perform autonomously : a perception-action loop with screenshots and DOM parsing handles pages that break brittle scripts. and which should always require approval.


• The observability, audit, and enterprise safeguards you’d expect before running this in production.


• The examples, samples, and tutorials that would help you get started fastest.

Responsible and secure by design

Because BAT lets an agent take real actions on live websites : isolated sessions, Live View, Take Control, and observability for reliability and audit. sometimes with credentials : scope credentials carefully and add your own mitigations; the tool is powerful and in preview. governance is a first-class part of the conversation, not a footnote. Isolated per-session sandboxes, Live View, Take Control human-in-the-loop, and built-in observability are there so you can see, pause, and audit what an agent does. Bring your trust concerns, required guardrails, and governance requirements: they directly shape the roadmap.

Note: the Browser Automation tool is in preview; APIs and capabilities may change, and it isn’t recommended for production workloads yet.

Key takeaways

• Browser Automation lets Foundry agents complete real web workflows : this round table feeds directly into the engineering and product teams. navigation, data gathering, form filling, research , and arrive on time for the tech check. via an MCP tool powered by Playwright Workspaces.


• Agents reason, not just replay , and browser-capable agents are how we cross it. a perception–action loop with screenshots and DOM parsing handles pages that break brittle scripts.


• Oversight is built in: isolated sessions, Live View, Take Control, and observability for reliability and audit.


• Responsibility is shared: scope credentials carefully and add your own mitigations; the tool is powerful and in preview.


• Your feedback shapes the product: this round table feeds directly into the engineering and product teams.

Save your spot

1. Add it to your calendar: 22 July 2026, 2:30 PM BST / 7:00 PM IST, and arrive on time for the tech check.


2. Join the community: https://aka.ms/foundry/discord


3. Prep with the sample: explore the browser automation sample in foundry-samples.


4. Read the docs: Automate browser tasks with Foundry agents and the hosted-agent quickstart.


5. Event registration link 
   

   https://discord.gg/Z8JZsrP5P5?event=1527676149264679013 

   
   

The last mile of automation still runs in a browser, and browser-capable agents are how we cross it. Come tell us what you’d automate, where you’d draw the line, and what you’d need to trust it in production. See you on 22 July.