AI Agents7 min read

Recent AI Agent Launches: Access Guide, November 2026

Published and checked

ℹ

Editorial note: Some links in this article are affiliate links — we may earn a commission if you sign up, at no extra cost to you. Research guides are distinguished from documented firsthand reviews in our editorial standards. Read our editorial standards →

Illustration accompanying an AI agent research article

Checked November 13, 2026. This selected roundup covers announcements from October 19 through November 10. It separates an agent runtime, a browser tool and a model update; it is not an exhaustive release calendar or a claim that every feature is available to every account.

The immediate buying question is which layer changed and what you can actually test. A preview harness, a generally available browser tool and a new underlying model do not have the same setup, cost or operational boundary. For the earlier period, see the May 2026 historical analysis.

Release and Access Snapshot

Selected developments; source status checked November 13, 2026
Announcement dateDevelopmentStatus stated by the sourceFirst access check
October 19Microsoft Foundry Browser AutomationGenerally available toolPlaywright Workspace, Foundry connection and the relevant permissions
October 28Copilot harness in Foundry Agent ServicePublic preview; hosted-agent resilience announced separately as GASupported model, region, tool configuration and preview limits
November 10Google Gemini 4 CarbonModel availability with access requirements; Antigravity /swarm described as a previewPaid API access or the stated product subscription; distinguish the model from /swarm

The November 5 Microsoft Foundry roundup recaps October developments. Its publication date is not a new launch date for each feature. This table deliberately records the underlying announcement dates.

Copilot Harness: Evaluate the Preview Boundary

Microsoft’s October 28 announcement describes a managed agent loop and sandbox, supported tools, reusable skills and source-controlled definitions. It labels the harness a public preview, while separately announcing general availability for long-running resilience in hosted agents.

The announcement says harness runtime billing begins December 1, 2026; model inference and separately billed tools still incur their applicable charges before then. A temporary runtime waiver is not a free end-to-end agent.

Our evaluation suggestion: start with a read-only research brief from approved documents. Check whether it can recover from a missing file and tell the reviewer what it could not establish. Use a hosted or custom approach only when you need control that the selected managed environment does not provide. That is a workload decision, not a declaration that preview software is unsuitable for every pilot.

Browser Automation: GA Does Not Remove Setup Work

Microsoft’s October 19 Browser Automation announcement describes a managed browser through Playwright Workspaces, integration with Foundry agents and oversight through Live View and Take control. It states that the workspace is a separately provisioned Azure resource. The product status is GA; your private application’s readiness is a separate question.

Our evaluation suggestion: use a staging portal with test records. Ask for a draft summary, then inspect the saved evidence. Leave form submission outside the first trial. Once navigation is reliable, test one explicitly approved state-changing action and verify both the intended record and the absence of duplicate actions.

A public demo cannot establish that your tenant configuration, private network path or sign-in method is supported. Record the exact site and account context in the pilot notes.

Gemini 4 Carbon: A Model Update Is Not a Complete Agent

Google’s November 10 announcement makes Carbon available through specified developer, enterprise and consumer access paths. It separately describes /swarm in Google Antigravity as a preview. Do not equate model availability with unrestricted access to every agent feature.

The announcement gives introductory public-model API pricing of $2 per million input tokens and $10 per million output tokens, with $4/$20 pricing after the introductory period expires November 30. These are USD token rates stated by Google, not the total cost of an agent workflow. Check the current rate before running a paid trial.

Our evaluation suggestion: retain the same prompt, documents and pass criteria when testing an underlying model change. Compare accepted results and correction time; do not attribute every improvement in an agent to the model if the harness, retrieval or tools changed as well.

An Original Pilot Brief You Can Reuse

Fictional test design; not an executed benchmark. Create three short documents describing a sample software plan, its price and an exception. Ask the agent for a one-page comparison that cites the source file for every material claim. Include one conflicting date and one missing feature so an unsupported confident answer cannot pass accidentally.

A small readiness trial
StepWhat to testEvidence to keep
Confirm accessOpen the required feature in your actual region and accountStatus, check date and relevant plan
Run a bounded taskProduce the same comparison from approved sample filesPrompt, source versions and output
Introduce an exceptionRemove a required source or add a conflicting dateUncertainty handling and reviewer decision
Review actionsKeep writes disabled until the read-only task passesAllowed actions and approval owner
Repeat and measureRepeat after one controlled changeAccepted outputs, retries, time and usage costs

Pass the document task only when the comparison identifies the conflict, avoids inventing the missing feature and cites the supplied files. That does not establish safe unattended operation; it establishes a narrower ability to complete this reviewed task.

Download the blank agent-launch readiness sheet (CSV) to record access, quotas, billing dates, service costs and approval ownership. The sheet contains headings, not scored vendor results.

Check Region, Currency and Total Cost

For US, Canadian and European teams, check availability for the specific product surface and account. A region mentioned in a launch post does not establish every data-location or contractual requirement your organization may have. Obtain the relevant provider documentation or agreement for those decisions.

Keep subscription access, model tokens, browser/workspace usage, storage and reviewer time in separate rows. Record the billing currency and term. For a credit-based product, a credit is not automatically equivalent to one token or one browser action.

Use LLM Pulse model search to shortlist underlying model records, then evaluate the complete workflow. A model comparison page cannot determine the availability or reliability of a separate browser runtime.

Frequently Asked Questions

Is this a list of launches from this week?

No. It is a dated selection of October–November developments checked November 13. Announcement, recap and check dates are kept separate.

Does general availability guarantee access in my account?

No. Verify the relevant region, account, configuration, quota and purchase requirements. We did not sign into provider accounts to test entitlement.

Are the vendors ranked by performance here?

No. These are different layers of an agent stack. No cross-vendor production benchmark was run.

Primary Sources and Coverage Limits

Sources checked November 13, 2026. Vendor documentation establishes advertised features and access, not independently measured performance. Examples and evaluation rules below are editorial guidance.

No paid product trial or production agent deployment was run for this article. Editorial standards · Report a correction.

This article uses official announcements for stated release facts and labels our own pilot suggestions separately. It does not repeat vendor benchmark rankings as independently verified findings. Later launches or access changes require another dated review.

Tags

ai agent product launch may 2026enterprise ai agents 2026microsoft copilot pricing 2026google io 2026 ai agentsonetrust ai inventoryai agent governancecopilot chat paid tierai agents production deploymentai agent tools 2026best ai agents enterpriseai workflow automation 2026ai agent limitationsgoogle ai agents availabilityai agent compliance

Written by

Sourabh Gupta

Sourabh Gupta

Author at Teach AI Tools

Sourabh Gupta is responsible for Teach AI Tools. Research guides and documented firsthand reviews are identified separately. An author profile is not evidence of testing; look for sources, test records, and limitations in the article.

Full bio & editorial process →

Related Articles