A team reviews AI workflow, security, and performance dashboards in a high-tech operations center.
Dominic Kent's UC Today guest column asks a question every IT shop should be asking right now: what if Microsoft's agent-first vision of work doesn't unfold the way Redmond says it will? It is an opinion piece, and Kent is careful to say he isn't claiming Microsoft is wrong. He argues that experimenting with AI is sensible, but rebuilding a company around one vendor's forecast is a different bet.

The column landed in the middle of Ignite 2026. That gives us a rare chance to hold the skeptic's questions up against what Microsoft actually shipped, and what status each feature has.

A team reviews AI workflow, security, and performance dashboards in a high-tech operations center. The argument: trying Copilot versus betting the company on it​

Kent separates two levels of commitment. In the first, a business rolls out Copilot to draft documents, summarize meetings and automate repetitive tasks. If it disappoints, staff go back to their old ways.

In the second, the business rebuilds its processes around autonomous agents. It invests in integrations, governance and training, and it changes job responsibilities. It also expects employees to reach their applications through a conversational, Copilot-first interface. At that point the company is building its operating model on Microsoft's prediction of how work will happen.

He supports the point with a comparison. The metaverse, VR-headset meetings and avatars were pitched as the future of collaboration, yet most businesses still meet much as they did before. That is his framing, not measured adoption data, and it's best read as an analogy rather than a statistic.

Eight what-ifs, translated into admin questions​

Kent raises eight concerns. Here is each one, with the evaluation question it implies for an IT team.

  • Interface fit. Is chat really faster than clicking? Kent points to Excel, PowerPoint and Dynamics, where direct manipulation can beat describing a task. His risks are misread instructions, hidden changes and harder troubleshooting. Ask: can employees still reach the apps directly, or will that be mandated away?
  • Skill retention. If agents build the financial model, the proposal or the project plan, juniors may never learn how. Ask: could the organization still operate during an outage or a platform change?
  • Action risk. A wrong answer from an assistant can be ignored. A wrong action from an agent can send an email, change a record or approve a transaction, and it can repeat before anyone notices. Ask: which actions require human approval?
  • Accountability. Is the person who instructed the agent responsible? The manager who approved it? IT? The process owner? The vendor? Kent says a named owner doesn't necessarily know what the agent did or why.
  • Economics. Licensing, consumption charges, integration, governance and human review all add up. His hypothetical contrasts 30 minutes saved with 20 minutes spent checking. It is an illustration, not a measurement. Ask: does the deployment improve the business, rather than just raising agent counts?
  • Resilience. Kent asks what happens when the agentic workforce "goes dark", for example during a Microsoft 365 outage. He documents no specific incident in the piece. Ask: what is the manual fallback?
  • Vendor change. Architectures, licensing and products change. Kent frames this as a dependency risk, not a prediction that any product will be retired.
  • Business purpose. Do customers care whether an agent handled their inquiry? Technology should serve the business, not the other way around.

His conclusion is not to avoid Copilot. It is to avoid irreversible dependence on one vision.

What Microsoft actually announced: Agentic Users in Teams​

Kent's accountability section hinges on Ignite news. Agentic Users, in public preview, let agents join a team under their own identity, with an accountable owner and admin controls. Microsoft's Microsoft 365 Developer Blog, published November 17, 2026, adds the detail:

  • Agents that act as themselves get their own Microsoft Entra accounts. Their messages and actions appear under the agent's name, not a person's account or a shared app.
  • People can find an agent by name, add it to chats, channels and meetings, and @mention it in Teams and across Microsoft 365 apps.
  • The agent's profile identifies its owner, and its actions are attributed to it in audit logs.
  • Microsoft says not every agent needs its own identity. Application-style agents and personal assistants may use different identity models.
  • Administrators approve the agent, define what it can access, review its activity in Microsoft Agent 365 down to individual tool calls, and disable it when needed.

The same post lists six other Teams updates. Threaded replies, agent plans and steps, HTML widgets and socket mode are in public preview. Streaming in group chats and channels and Extended Markdown are available now. Plans and steps are the most relevant to Kent's visibility worry, because completed steps remain available for review next to the final response. In Microsoft's own demo, the agent explains its recommendation and the decision to commit stays with the team.

The wider governance push​

Microsoft's Ignite Book of News frames Agent 365 and related Entra, Intune, Defender and Purview capabilities as giving organizations greater visibility and control over agents, their identities, tools, access and data. Some details worth knowing, with their status:

  • Management insights for Microsoft prebuilt agents and agents built on Microsoft platforms are generally available in the Microsoft 365 admin center without an Agent 365 license. Enterprise-grade management, security and governance capabilities still require an Agent 365 or ME7 license.
  • An action center for inactive agents and agents without owners is expected in private preview in December.
  • Dynamic Agent Risk is in preview. It combines identity, access, data, endpoint, cloud, network and posture signals into a continuously updated assessment. Partner signals from Check Point, Cyera, Darktrace and Zenity are due to arrive through Microsoft Sentinel.
  • Microsoft's Entra announcements describe discovery of local agents. Discovery expands by 35 local agent types, to 60 in total, including Claude Code, OpenAI Codex, and GitHub Copilot, and discovered agents receive a Microsoft Entra Agent ID and can be governed with Conditional Access. That is notable for Kent's lock-in worry, because it extends governance to some non-Microsoft agents.
  • Third-party coverage of the Entra release says the default lifecycle policy disables agent identities inactive for 60 days, and that automated lifecycle management is generally available. This comes from the aguidetocloud and WindowsForum coverage of Ignite, not from Microsoft's own page, so verify it in your tenant.

Treat these as Microsoft's product descriptions. None of them independently proves that the risk scoring is accurate or that the controls prevent bad agent behavior.

Identity is not accountability​

Kent's sharpest point is that giving an agent an identity and an owner is not the same as establishing accountability. The Microsoft materials back the narrower claim: they describe identity, ownership, audit attribution and revocation. They don't claim to settle liability, and they don't claim every agent needs its own identity.

A named owner gives you somewhere to send the pager. It does not tell you whether the agent's decision was reasonable. UC Today's own Ignite coverage puts it plainly: if agents get @mentioned in Teams, IT needs a named human behind every one.

Reading the column critically​

Kent works closely with vendors such as Nextiva and Cisco, and his consultancy serves technology companies. That's disclosed in his bio, and it doesn't invalidate his questions. Still, those relationships are the context in which a column urging vendor diversification should be read.

There is also a counterargument. Microsoft isn't asking for an all-or-nothing switch. Per the Book of News, Office in Copilot is generally available, and Copilot's Home and Auto features are rolling out. Nothing in the evidence suggests direct access to Word or Excel is being removed. The "mandated Copilot-first" scenario is Kent's hypothetical, not a Microsoft announcement.

A practical checklist for admins​

Here is how to turn the commentary into a pilot plan:

  1. Keep a direct path. Confirm users can still work in Word, Excel and line-of-business apps without the agent layer.
  2. Define approval gates. List which agent actions, such as sending messages, changing records or approving payments, need a human sign-off.
  3. Assign real owners. Treat ownership as a duty with training and review time, not a name in a profile. Add agent ownership to offboarding.
  4. Check the license path. Confirm which governance features need Agent 365 or ME7 before you plan around them.
  5. Respect the preview label. Agentic Users, threaded replies, plans and steps, and HTML widgets are public preview. Don't build production dependency on them yet.
  6. Cost the whole loop. Include review time, maintenance and consumption charges, and measure business outcomes rather than the number of agents.
  7. Plan the dark day. Write the manual fallback for when agents or the underlying service are unavailable.
  8. Keep exits open. Keep process documentation, skills and integrations portable so a change in architecture or pricing doesn't strand you.

Bottom line​

Kent doesn't offer evidence that Microsoft's vision will fail, and he says so. What he offers is a disciplined set of questions. Microsoft's Ignite announcements answer some of them with identity, audit and admin controls, mostly still in preview. They leave the harder ones, such as policy, oversight, costs and fallback, with the customer. Adopt the tools, pilot them, and keep a way out.

 

References

  1. What If Microsoft Has Got It Wrong? Don't Bet on One Vision - UC Today UC Today 2026-11-20T15:57:03.346000+00:00
  2. Microsoft Ignite 2026 news.microsoft.com
  3. New Capabilities for Building Digital Teammates in Microsoft Teams - Microsoft 365 Developer Blog devblogs.microsoft.com