Dirty Static // The Verdict
The most governed personal agent yet from a big platform, worth a small pilot in Frontier, not yet a product most teams can buy or budget.
Best forMicrosoft 365 teams in Frontier who have recurring follow up work, a named manager for the agent, and a spending limit set before it starts
What works
- It keeps working without a new prompt, with its own identity, memory, workspace and tools inside your tenant
- Other people can work with it in Teams and Outlook and mention it like a colleague, so it is not trapped in one private chat
- Its actions are logged under its own Entra Agent ID, IT can scope its access and block it, and it can check with you before key actions
- The one detailed daily account shows a sensible boundary in practice, with drafts prepared for review instead of messages sent on its own
What does not
- It is rolling out to Frontier only from November 17, 2026, with no general availability date and eligibility gates most tenants do not yet meet
- It runs on usage based billing in Copilot Credits on top of the Copilot licence, and Microsoft has not published how many credits Autopilot work uses
- There is no independent testing record yet, only Microsoft launch material and one detailed account from inside Microsoft
- It shares a name with Windows Autopilot, the device setup tool, which will confuse staff, search results and support tickets
Research review: based on published testing, specifications, and documentation, not hands on testing.
Most Copilot use still ends when you close the chat. You ask, it answers, and the follow up work goes back on your list. Autopilot is Microsoft attempt to break that pattern. It is a personal agent that keeps working between prompts, watches the places where your work already happens, and comes back with work ready for your attention.
Microsoft began rolling Autopilot out to Frontier customers at Ignite on November 17, 2026. That timing matters for how to read every claim in this review. This is a research review, not a test. Dirty Static has not run Autopilot, no independent testing record exists yet, and the richest account of daily use comes from Microsoft itself. What we can do now is separate what Microsoft has actually documented from what it hopes the product becomes.
What It Is
Autopilot lives in your tenant and has its own identity, memory and tools, per the Copilot blog post for Ignite. It shows up in Teams, Outlook, chats, channels and documents, so colleagues can mention it like a colleague. Because it runs in the cloud, the work does not depend on keeping your laptop open.
The history helps explain the shape. Microsoft introduced Scout in June 2026 as its first always on personal agent. On September 25, 2026 it renamed Scout to Autopilot and made it one of the three parts of the rebuilt Copilot, beside Home and Code. Our report on Office inside the Copilot app covers the wider Ignite launch and where Autopilot sat in it on the day.
- June 2, 2026Microsoft introduces Scout, its first always on personal agent
- September 25, 2026Scout is renamed Autopilot in the rebuilt Copilot with Home, Code and Autopilot
- November 17, 2026At Ignite, Microsoft says Autopilot is beginning to roll out to Frontier customers
Features and Specifications
Specifications
- Product
- Copilot Autopilot
- Maker
- Microsoft
- Former name
- Scout
- Status
- Rolling out to Frontier customers from November 17, 2026
- Where it runs
- In the cloud, inside the Microsoft 365 tenant
- Identity
- Its own Microsoft Entra Agent ID, with a named manager
- What you give it
- A name, a role and a goal
- Where it appears
- Teams, Outlook, chats, channels and documents
- Context source
- Work IQ, per the Ignite post, grounded in mail, meetings, files and chats
- Billing
- Usage based billing in Copilot Credits, on top of the Copilot licence
Microsoft gives one full example of the work it has in mind. Autopilot can set up and manage a full supplier review process, building the schedule and the workback plan, then handling prep, meetings and follow ups on its own, right down to reaching out to stakeholders for updates. Read that as a Microsoft example of intended scope, not as a measured result. No customer is named for that example in the Ignite post, and no completion rate is published.
The controls are part of the specification, not an extra. Each Autopilot has its own Entra Agent ID and a named manager. IT can scope its access and block it at any time. Its sign ins and actions are logged under its own name, so security teams have a clear audit trail. Microsoft also says you decide what work Autopilot takes on and that it can check with you before key actions. That last line will decide whether teams trust it in practice, because the difference between preparing a reply and sending it is the difference between help and trouble.
The useful pattern in the early account is drafts waiting for review, not messages already sent. Illustration: Dirty Static
Real World Experience
The only detailed daily account we have is Microsoft own. In a Tech Community post published for Ignite, the author describes working with an Autopilot named Sebastian in Teams and Outlook. Sebastian helps prepare for the day, works through conversations that need a reply, and coordinates meetings. Some jobs are one time requests. Others run on a routine without a new prompt.
Two details in that account are worth more than the launch language. First, Sebastian sends a Sunday Reset by email, and the author can reply to that email and continue the conversation. Other people can also schedule time by talking to Sebastian in Teams, using scheduling preferences the author has already set, and work can be delegated to Sebastian inside group chats and channels. That is what its own identity buys in practice. The agent is reachable where the work is, by the people involved, without the owner relaying every message.
Second, the boundary held in the example that matters most. While out of the office, the author asked Sebastian to help with unanswered Teams conversations. Sebastian prepared 40 reply drafts, grouped them by priority, and put them in a private editable queue for review. The replies stayed as drafts. The author remained responsible for what would be sent. Microsoft post is open about why that distinction matters, noting that preparing a reply and sending it are different things, and that finding a time and confirming a meeting are different things.
Treat the rest of that post with care. A longer example about getting to Ignite and back, covering flights, hotels, registration, receipts and the expense work after the trip, is presented as ambition. The author says day to day work with Sebastian is helping Microsoft understand what it takes to make that ambition dependable, and that the team is still building and learning. That honesty is welcome. It also means no reader should quote the travel story as proof that Autopilot can already run a trip end to end.
There is no independent account to set beside this one yet. Autopilot only began its Frontier rollout on November 17, so outside testers have had no real time with it, and Microsoft has published no benchmark, no task completion rate and no reliability figure for Autopilot itself. Anyone who needs those numbers before a rollout has to wait, or run a pilot small enough to produce their own.
What Works
Continuity is the real gain. A normal chat starts over each time unless you rebuild the context. Autopilot carries memory and context forward, builds on what it learned last week, and returns to a project days later without you explaining the whole story again. For recurring work, the kind that is ordinary and easy to drop, checking, coordinating and following up, that continuity is worth more than a smarter single answer.
The identity model is the second strength, and it is the right design. An agent that acts as you, under your account, blurs who did what. An agent with its own identity, its own log and a named manager can be scoped, watched and blocked like any other account in the tenant. Microsoft is also extending governance through Agent 365, which the Ignite post describes as the place for visibility, access controls and spend controls across agents. That does not make Autopilot safe by default. It makes it governable, which is the condition a serious firm needs before safety is even a discussion.
The draft first pattern in the Sebastian account is the third strength. Bringing work to the point where a person can review it and decide quickly is a better near term promise than full autonomy. A morning briefing that arrives without a prompt, a queue of drafts sorted by priority, a meeting change proposed with context, those are gains a team can use this month, while the grander talk of an agent that manages a whole process on its own still needs proof.
What Does Not
Access is the first problem, and it is not close. Autopilot is not generally available. The Tech Community post says that to be considered eligible, customers need to be enrolled in Frontier, with a Copilot licence, usage based billing enabled, and the required management and security capabilities in place. Reporting on the rollout adds Agent 365 to that list and notes that Autopilot is off by default and must be enabled by an admin, starting with a pilot group. If your tenant is not in Frontier, there is nothing to evaluate yet beyond the paperwork.
Reliability is the second problem, and Microsoft own words frame it. The Sebastian post says reliability matters as much as capability, and that completing a task includes knowing which steps actually happened. That is exactly right, and it is an admission that the hard part is unfinished. An agent that watches channels and follows up with stakeholders can create work as fast as it clears work when it misreads a thread, chases the wrong person or acts on stale context. The controls let IT block it. They do not tell a manager how often it will need blocking.
The third problem is scope creep in the pitch. Supplier reviews, travel, meeting coordination, inbox triage and project management are all in the launch material, but the documented daily use is narrower, briefings, drafts, scheduling help and follow ups. Teams that buy the broad pitch and assign broad goals on day one will learn the limits in public, in front of colleagues and customers. The product rewards the opposite approach, one responsibility, clear boundaries, review of the output, then more scope only after it earns it.
Competition
| Option | What it is | How it differs from Autopilot |
|---|---|---|
| Copilot Autopilot | Persistent personal agent with its own identity in your tenant | Keeps a responsibility over time, reachable by colleagues in Teams and Outlook, billed by usage |
| Copilot Cowork | The delegation surface inside Copilot Home | Takes a task and returns finished work for review, without the standing personal identity and ongoing watch that define Autopilot |
| Claude Cowork | Anthropic agent for files, folders and connected tools | Works across the files and tools you choose and returns work for review, as covered in our Claude in Chrome and Cowork review, but it does not live in your Microsoft tenant under an Entra identity your IT team governs |
The choice is mostly about where your work already lives and who must govern the agent. A Microsoft 365 firm that needs an audit trail under its own directory, with the agent reachable in Teams, has no real rival to Autopilot in that exact shape. A team whose work lives in files and browser tabs, and that already pays for Claude, may get more usable agent hours today from a product that is already generally available, even without the tenant identity. And a team with no Frontier place should treat both rows as planning, not purchasing, because the Microsoft row cannot be bought in the normal way yet.
Final Verdict
Pilot Autopilot if you are in Frontier, you have one recurring responsibility with a clear owner, and your admin will set a spending limit and approval boundaries before the first goal is assigned. Start with drafts and briefings. Read the logs. Count how often it prepares the right thing, not how impressive the demo language sounds. That is the only honest test available right now, and Microsoft own account suggests the product can pass it on narrow work.
Wait if you need a price, a general availability date, or proof from anyone outside Microsoft. All three are missing. Autopilot is the most serious attempt yet by a big platform to give a personal agent a real identity, a real manager and a real audit trail inside the tools a firm already runs. That design deserves attention. It does not yet deserve a broad rollout, a large budget, or your trust on work where a wrong send, a wrong booking or a wrong follow up would be hard to undo.
Facts in this review were checked on November 18, 2026 against the sources below. This is a research review. No hands on testing was performed. Company claims and Microsoft own daily account are labelled as such where they appear, and final Autopilot pricing had not been published at the time of checking.
Sources
- Copilot at Microsoft Ignite, Microsoft Copilot Blog, November 17, 2026
- Autopilot, meet your trusted personal teammate, Microsoft Tech Community, November 17, 2026
- Microsoft Copilot Autopilot, the always on AI agent for Microsoft 365, Rob Quickenden, November 2026, on prerequisites, Frontier rollout and usage based billing
- Microsoft makes Office in Copilot generally available as Autopilot rolls out, TokenPost, November 18, 2026, on Frontier eligibility and final pricing to come
- Copilot Credits explained, Solv Systems, on the licence and credit model from Microsoft September 2026 pricing posts
- Hero and body images: Dirty Static, generated in house style
END OF TRANSMISSION
Stay curious. Question everything.


