Microsoft Ignite 2026: Agent Identity, AI Cost Control and Agentic Security

Microsoft spent Ignite 2026 on agents, and the open question across more than 180 posts was who controls them. Agents now get their own identities, audit trails, rollback and cost dashboards. Building an agent got easier, while running a few hundred of them is where the work sits now. Security shifted too, with agents treated as identities, insider risk and defenders at once.
Date
November 18, 2026
Special - Digital
Thomas Allgeyer

Methodology: Every two weeks we collect most relevant posts on LinkedIn for selected topics and create an overall summary only based on these posts. If you´re interested in the single posts behind, you can find them here: https://linktr.ee/thomasallgeyer. Have a great read!

Listen to our podcast

If you prefer listening, check out our podcast summarizing the most relevant insights from Microsoft Ignite 2026:

Agents Get Identities

  • Each Autopilot gets its own Entra Agent ID and a named human manager, and IT can scope or block it at any time.
  • Every Autopilot action is logged under the agent's own name, which closes the gap of automations running on a founder's login.
  • Entra disables and then deletes agent identities that have no sponsor or no recent activity, and this is generally available.
  • Security teams now frame the agent as an identity that needs an owner, the right permissions and an off switch.
  • Entra Conditional Access separates the user from the agent acting for them, and an MCP firewall was added.

Governance Moves Into Platform

  • Governance is leaving the policy document and moving into the platform, but controls only enforce what someone decided first.
  • Customer questions have shifted from whether to allow agents to how to govern them with content guardrails.
  • Agent 365 now covers more connected platforms, including Anthropic, Databricks, OpenAI and Salesforce, plus multi-tenant management.
  • Agent 365 already manages more than 50 million agents, according to the keynote as relayed in the posts.
  • Governance questions now include ownership after a maker leaves, approval of sensitive SharePoint access and which team drives usage-based spend.

Cost and Value Visibility

  • Agent 365 Business Analytics puts AI spend and value in one view, framed as FinOps for AI.
  • It shows spend by service, model, agent and user, with AI activity mapped to more than 6,000 business use cases.
  • Leaders can judge value through success rate, user satisfaction and assisted hours, and use forecasting, what-if scenarios, limits and alerts.
  • Falling cost of creating agents raises the cost of managing them, which makes orchestration the real enterprise challenge.
  • One demo showed a fine-tuned agent that was more accurate and 98% cheaper than other builds of the same agent.
  • Work IQ billing is opaque, since the Azure invoice does not show the name Work IQ.

Rollback and Trust

  • Agent Rewind, in private preview, restores Microsoft 365 content after unwanted agent actions, including large-scale incidents.
  • The AI Transparency Initiative records what agents were authorized to do, which model they used and what they touched.
  • The initiative works with more than 40 organizations, including Accenture, Adobe, KPMG, Mastercard and Volvo Cars.
  • Agent time savings can be offset when someone else spends hours checking and fixing the work.
  • More autonomy demands stronger human ownership, verification and accountability, with approval points built into rollout.

From Copilot to Autopilot

  • The progression runs from Copilot to Cowork to Autopilot: ask, delegate, then have AI keep work moving without a prompt.
  • Autopilot is an always-on agent that watches Teams channels, follows up on emails and runs recurring work, rolling out to Frontier customers.
  • Copilot is positioned as the new operating system for work, with Word, Excel and PowerPoint inside Copilot chat.
  • Copilot Workspaces gives people and agents a shared space, and the enhanced experience is in Frontier.
  • Rollout timelines differ widely, with Home and Managed Runtime available now and Autopilot, Workspaces and Today earlier in preview.

Copilotforce and Integration

  • Copilotforce brings Salesforce customer context and sales workflows into Microsoft 365 Copilot, Teams and Fabric.
  • Salesforce in Copilot is in preview with 38+ prebuilt sales skills through authenticated Salesforce MCP access.
  • Existing Salesforce permissions keep governing what a seller can see and do, so governance stays with the system of record.
  • A two-way zero-copy integration between Salesforce Data 360 and Fabric lets teams analyze data without duplicate datasets.
  • Dynamics 365 plugins add 100+ prebuilt skills, and business staff can run sales, service, finance and purchasing tasks by conversation.

Agentic Security Operations

  • Microsoft Perception is generally available, with red, blue and green team agents working together while humans keep control of critical decisions.
  • The Triage Agent can handle about 80% of alerts in a typical Defender queue.
  • AI shortens exploit timelines: one researcher turned critical PHP flaws into a working attack in about a week, work that once took a specialist team months.
  • About 41 million internet-facing services ran affected PHP versions on the day of disclosure.
  • The Integrated SOC in Defender brings SIEM capabilities in, with case management, playbooks, workbooks and UEBA generally available.
  • Agent 365 adds Dynamic Agent Risk in preview, giving one live risk score per agent.

Models and Developer Practice

  • Seven new MAI models make cheaper, tuned models a real option for enterprise workloads.
  • Model choice is getting easier, while dependence shifts to the identity, context, governance and billing layers of one supplier.
  • Copilot Studio is repositioned as an Agent DevOps platform, with Dev, Test and Prod pipelines, version control and approval gates.
  • Copilot Managed Runtime is generally available, with Microsoft-operated hosting, Entra identity and governed data access.
  • Foundry tracing and rubric-based evaluators let teams define what good looks like and score agents against it.

Thanks to Pablo Carmona, Chris Peña, Mike Yeh, Melisa Sever, Diego Plaza, Tavon Giaquinta, Ralf Thomas, Rahul Sapre, Daniel Würsch, Abdul Khadja Alaoudine, Arun Sabharwal, Josipa P., Siva Polisetty, Gonzalo Bernal Morales, Ruben Lucas, Vishal Goyel, Bjoern Hirtenjohann, Jan-Frederik Kobus, Mike Primeau, Arnav Pandya, Koji Kusunoki, Sneha Daggubati, Angela Schnellert, Abhishek S., ISAAC FORD-WILSON, Gina Montgomery, Taras Kloba, Nicole Chesworth, MBA, Naveen Gaur, Brij Pandey, Ken Osterhaus, Cherie Gartner, Tobias Regnström, Karol J., Sesha Mani, Nati Papirovitch, Aleš Holeček, Ricard Farrés, Simon M., Petri Pukkanen, Emanuele Dicuonzo, Silvia Rodenas Vaquero, Jack Rowbotham, Remco Ploeg, Teresa B. Cyrus, MVP, Itzhak Janach, Rafsan Huseynov, Arjen Furster, Prashob Jayaprakash Narayanan, Lydia Makau, Rob Lefferts, Haoyi Yin, Kartik Datwani, Nathalie van Rooij, Dominic Kent, Herain Oberoi, JP Bourget, Mark Vigoroso, MBA, Jean Marie Sadio, Nursultan Kireyev, Arindam Choudhury, Thiago Ribeiro, Sean Parham, Gerson Viergutz, Emad Adel, SaiSree Marupudi, Baki ÇUBUK, Lisa Crosbie, Kaushal Todi, Francesco De Simone, Lukas Wagner, Martijn Furster, Khalid IBNELBACHYR, Ilya Lushnikov, Lee Mather, ShiSh S., Dave Galy, MBA ☁️, Luiz Pirollo Neto, René Raeber, Vincent Sterken, Tom Jackson, Kiran Arunkumar Babu, Phillipe Chaves de Lima, Brenner Melo, Robert Paradise, Melissa Ambers, LSS, PMP, PR, FCGI, Houtan Rasouli, Cynthia Akiotu, Tomi Koskinen, Graham O’Sullivan, Vladimirs Surajevs MCSE MCSA, Michael Böbner, Tariq Ali, Vikram Solanki, Xing Mei, Michael Altenberger, Tejas Shah, David J. Chavez, Dinesh Chandrasekhar, Thirumalesh Thatipamula, Robert Bruckner, Jorge Balderas, Scott Laswick, Michael Morten Sonne, Ivana Tilca, Jean-Christophe Guyard, Kate Macartney, Waqas Saeed, Eric Chaniot, Charlie Gough, Vepa Oraz, 🛡️ Seyfallah Tagrerout☁ [MVP and RD], Oliver Lindsay, Jia Ashitta, Paulo de Tarso Carvalho, Anton M., Vijayan R., Dennis Padia, Murat Everekli, Christoffer Andenmatten, Paul Snider, Thulya Bezawada, Antonio Formato, 김영욱, Zoe R., Micah Heaton, Brian Swiger, Gopi Thangavel Ph.D, CCISO, CISM, PMP, EPLCM, ITIL, Mohammad Ali, Dennie Kooijman, Dux Raymond Sy, David Geddam, Laide Majiyagbe, David Rosenthal, Joshua Hoffman, Thomas Poppelgaard, Jessica Geiger, Alex Sharp, Guruprasad AM, Marko Lauren, Meena J Gowdar, Ugur Koc, Elkhan Y., Samuel GASTON-RAOUL, Ananya Ghosh Chowdhury, Luuk Ros, Saurabh Rawat, Vishal Vasu, David Alonso Dominguez, Michael Dawson, Reynold D'Sa, Megan Gremmell, Karthik Reddy Yanamala, Wangui McKelvey, MEGHANA VEMPATAPU, Rahul Neel Mani, Dushan Tharmal, Dona Sarkar, Tushar Gupta, Rajnish Malik, Sly Gittens, Markus Ikonen, Vyom Nagrani, Lee Harrison, Mihir Jhaveri (PMP, F.IoD), Aidan Forman, Andreas Selevik, Maria S., Richard Yang, Timo Haslberger, Rowan Shield, Femke Cornelissen ✨, Yash Lara, Danny Mwiinga, Thomas Blöchl, Eva Benn, ADHARSH KUMAR BACHU, Arun Ulag, Christiaan Brinkhoff, Tina Schuchman, Ademola Isimeme (AI) Odewade, Kevin Griffin, Lauren Nordmann, Alvin George, Soumik M., Ravit Jain, Markus A. Montilla and Sumit Chauhan and everyone else who contributed insights to this edition. Find the full list of posts and voices on LinkedIn.

Subscribe to newsletter

Subscribe to receive the latest blog posts to your inbox every week.

Please confirm your GDPR consent to join our mailing list.
By subscribing you agree to with our Privacy Policy.
Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.

Frequently asked questions

What was the big shift at Microsoft Ignite 2026?

The shift was from building agents to controlling them. Agents now get their own identity, audit trail, rollback and cost view. Agent 365 already manages more than 50 million agents. Autopilot gets an Entra Agent ID and a named human manager, and IT can block it at any time.

What is Agent 365 Business Analytics?

It is a dedicated view that puts AI spend and business value together, framed as FinOps for AI. It shows spend by service, model, agent and user. AI activity maps to more than 6,000 business use cases. Leaders also get forecasting, what-if scenarios, spending policies, limits and alerts.

What is Microsoft Perception and why does it matter?

Microsoft Perception is a generally available security system where red, blue and green team agents detect, investigate and fix threats while humans keep control of critical decisions. Its Triage Agent can handle about 80% of alerts in a typical Defender queue. One researcher exploited critical PHP flaws in about a week, which once took months.

No items found.

Want to see the posts voices behind this summary?

This roundup brings you the Best of LinkedIn on Microsoft Ignite 2026:

‍