Microsoft Copilot New Features: AI Coworkers in 2026
Microsoft rebuilt Copilot around Home, Code and Autopilot, adding a persistent AI coworker with its own company-directory identity, natural-language app building, and new admin cost controls. Here is what business owners and IT managers must know before enabling it.
The Short Answer
On September 25, 2026, Microsoft unveiled the biggest redesign of its AI assistant since launch, and the Microsoft Copilot new features signal a decisive shift: Copilot is no longer just a chatbot that answers questions. Announced in an official blog post by Jared Spataro, Chief Marketing Officer for AI at Work, the overhaul rebuilds Copilot…
On September 25, 2026, Microsoft unveiled the biggest redesign of its AI assistant since launch, and the Microsoft Copilot new features signal a decisive shift: Copilot is no longer just a chatbot that answers questions. Announced in an official blog post by Jared Spataro, Chief Marketing Officer for AI at Work, the overhaul rebuilds Copilot around three pillars, Home, Code and Autopilot, and introduces a persistent digital coworker that carries its own identity inside your company directory. For business owners and IT managers in the UK, US and Pakistan, this changes how teams get work done, how agents are governed, and how AI spending is kept under control.
Microsoft Copilot New Features Explained: Home, Code and Autopilot
The redesign moves Copilot from a side panel you chat with to a work surface you delegate to. Everything rolls out first through Microsoft's Frontier early-access program, with broader availability following. Here is what each pillar actually does.
Home: a new landing experience with Office built in
Home becomes the starting point for work inside Copilot. It combines Chat and Cowork (delegated, multi-step work) in one place, and, most practically for office teams, Word, Excel and PowerPoint now run directly inside the Copilot experience. Drafting a proposal, building a spreadsheet model or laying out a deck no longer means switching between apps while Copilot watches from the sidebar.
Home also gains a proactive Today command center, entering private preview in October 2026. It works across your email, calendars, meetings and Teams threads, pulling together what needs attention and drafting follow-ups that are ready to send. Microsoft is also adding @Copilot inside Teams to a private preview, so you can summon Copilot directly inside team conversations.
Code: build apps, dashboards and automations from plain language
Code lets people who do not write code describe an app, tracker, dashboard, automation or workflow in natural language and have Copilot build it. It runs on the same underlying technology as GitHub Copilot and executes inside a sandboxed environment, with apps hostable securely within your own Microsoft tenant through the preview of Copilot Managed Runtime.
This is the feature with the biggest day-to-day impact for non-technical teams. A sales manager could describe a deal tracker, or an operations lead could describe an approval workflow, and get a working internal tool instead of another spreadsheet. That also means IT suddenly needs a policy for who can publish what inside the tenant, which is exactly the kind of tenant-level oversight many growing companies fold into their managed IT services arrangement.
Autopilot: the persistent digital coworker
Autopilot is the most agentic part of the announcement. It is a revamp of the "Scout" agent Microsoft introduced in June 2026, and Microsoft describes it as a persistent, proactive digital teammate. Unlike a chatbot, Autopilot keeps working on recurring tasks without waiting for a fresh prompt: monitoring channels, following up on threads, running regular work and resuming projects days later.
Two details matter for governance. First, Autopilot gets its own identity and workspace inside your company directory, so it shows up as a recognizable entity, not an invisible background process. Second, it carries configurable permissions that decide what it can read and do. This is the same direction the wider industry is moving in, as we explored in our breakdown of always-on AI agents for business, and it is the feature most likely to change how recurring office work gets handled in 2026.
What Changes for Your Team Day to Day
For a typical 20 to 200 person business, three practical shifts stand out:
- Less context switching. Because documents, spreadsheets and presentations live inside Copilot, routine office production (proposals, reports, client decks, financial models) stays in one flow, with the assistant aware of the document and the conversation together.
- Citizen-built tools. Code lowers the barrier so far that department heads will start building their own dashboards and automations. That is a productivity win, but without rules it becomes shadow IT. Set expectations early about which tools are personal, which are departmental, and which need IT sign-off.
- Agents that keep working while you sleep. Autopilot can chase an overdue invoice, summarize a channel that moved fast overnight, or keep a recurring report running on schedule. The question is not whether this is useful, it clearly is, but who approved the agent, what data it touches, and who is accountable when it acts.
If that accountability question made you pause, you are not alone. The statistics on AI adoption and governance suggest most organizations are deploying AI faster than they are governing it, which brings us to the numbers.
The Numbers Behind the Hype: Why Governance Cannot Wait
Copilot's new features land in workplaces where AI use is already widespread but governance is thin. Consider what recent surveys found:
| Finding | Source |
|---|---|
| About 35% of UK businesses with 10 or more employees were using at least one AI technology by June 2026, up from around 12% in late 2023. 55% of employees report using AI for work or education. | ONS analysis, via CV-Library |
| 61% of service firms in the New York Fed's region used AI in 2026, up from 40% in 2025, but the median share of staff actually using it was 17%. | New York Fed September survey, via HybridHero |
| 56% of US workers have used generative AI for work tasks, yet only 26% of organizations have a policy governing generative AI use, and 34% have no AI policy at all. | The Conference Board survey of nearly 1,100 US employees, via SQ Magazine |
| 53% of organizations had deployed AI agents by Q2 2026, and the share orchestrating multiple AI agents across workflows doubled from 9% to 18% in a single quarter. | KPMG, via SeoProfy |
| 86% of IT and security leaders anticipate AI agents will surpass their organization's security guardrails within the next year, while only 23% claim full visibility into the AI agents operating in their environments. | Rubrik Zero Labs survey of 1,600+ leaders, via MSSP Alert |
| 71% of organizations use AI tools that access core business systems such as Salesforce and SAP, but only 16% effectively govern that access. | 2026 CISO AI Risk Report, via Dataconomy |
| 86% of organizations experienced at least one AI-related incident in the past year, such as data exposure or unapproved employee AI use. | OneTrust 2026 AI-Ready Governance Report, via Daily Guardian |
The pattern is consistent: employees adopt AI quickly, agents are being deployed, and oversight lags behind. Microsoft clearly knows this is the barrier, because Annie Pearl, Corporate Vice President for Copilot Product, has said that security, compliance and governance concerns shaped how these new features were designed. That is reassuring, but design intent is not the same as a configured tenant. As we covered in our guide to AI agent security controls, the protection comes from how you configure and police agent access, not from the feature existing.
Microsoft Copilot New Features: A Governance Checklist Before You Enable Persistent Agents
Before you switch on Autopilot or let Code apps loose in your tenant, work through this checklist. It applies whether you have an in-house IT team or work with an external IT consulting partner.
- Inventory your agent identities. Autopilot gets its own directory identity, which is good, because you can then apply the same least-privilege principles you use for human accounts. Document who created each agent, who owns it, and what business purpose it serves.
- Scope permissions before deployment. Autopilot supports configurable permissions. Restrict which mailboxes, SharePoint sites, Teams channels, Dynamics 365 records and Power Platform connectors each agent can touch. A follow-up agent does not need finance's folders.
- Require approval for publishing Code apps. A dashboard built from a sentence is still an application. Decide which apps can be published tenant-wide, who reviews data access, and how long approvals stay valid.
- Turn on usage visibility. Microsoft's new controls give employees visibility into their own usage; make sure your admins get the equivalent view across teams, so one enthusiastic department cannot quietly burn through your Copilot Credits.
- Set spending policies early. The per-user license covers everyday AI use, but long-running agent work is usage-based. Define monthly guardrails per team or user group, and set alerts before limits are hit.
- Run a shadow-AI sweep first. With 82% of enterprises in one survey discovering agents their security team never knew about, assume you have unapproved tools running and find them before you bless the official ones.
- Define the rollback plan. Know how to revoke an agent's permissions and undo its recent actions without disrupting systems. Over 80% of leaders in the Rubrik research say agents currently need more manual oversight than they give back in efficiency, so plan for human review of external-facing actions.
- Train staff on what agents are. People will trust an agent with a company-directory identity. Teach them to verify agent outputs the same way they would verify a junior colleague's work, especially before anything is sent to customers.
Keeping Agent Spend Under Control
The pricing change inside this announcement deserves attention from anyone who signs the Microsoft bill. The model now splits in two: a per-user license for everyday AI assistance, and usage-based Copilot Credits for longer-running agentic work. That mirrors how the product actually gets used, but it also means an always-on agent can quietly generate usage-based spend while nobody is watching.
Microsoft's answer is new cost governance for admins: spending policies, Copilot Credits, model-family limits per user group, and usage visibility for employees. Treat these like you would cloud spend controls. Pilot with one team, watch the consumption numbers for a month, then roll out with per-group limits in place. Enterprises worried about runaway AI costs finally get the levers, but only if someone actually configures them.
Frequently Asked Questions
What are the Microsoft Copilot new features announced in September 2026?
The three pillars are Home, a new landing experience combining Chat and Cowork with Word, Excel and PowerPoint embedded directly in Copilot; Code, which builds apps, dashboards and automations from natural-language prompts using the same technology as GitHub Copilot in a sandboxed runtime; and Autopilot, a persistent digital coworker with its own company-directory identity and configurable permissions that keeps working on recurring tasks without fresh prompts. The announcement also adds the proactive Today command center in Home, @Copilot in Teams, and new admin cost-governance controls.
When will Home, Code and Autopilot be available to my business?
Microsoft is rolling them out gradually. Home and Code begin reaching users through the Frontier early-access program in the weeks after the September 25 announcement, while Autopilot moves into private preview around the end of September 2026. A broader preview of Code for Microsoft 365 Premium and Pro subscribers is planned later in 2026. If your tenant is not in Frontier, expect a wait, and use the time to prepare the governance checklist above.
Will Autopilot replace employees?
No, and that is not how Microsoft positions it. Autopilot handles recurring, well-defined work, monitoring threads, following up, keeping recurring tasks running, while people keep the judgment calls. Recent research puts the productivity angle in perspective: the Stanford AI Index 2026 measured gains of 14 to 15% in customer support and 26% in software development from AI assistance. The realistic outcome is fewer repetitive chores per employee, not fewer employees, provided you govern the agents properly.
How do permissions work for a persistent agent?
Autopilot receives its own identity in your company directory with configurable permissions, so you control exactly which mailboxes, sites, channels and business systems it can access. This is a meaningful step up from agents running on borrowed credentials, but it only helps if you scope those permissions narrowly and review them periodically, the same discipline you apply to human access rights.
What does usage-based billing mean for my AI budget?
Everyday Copilot use stays on the per-user license you know, while longer-running agent work is billed by usage through Copilot Credits. Admins get new spending policies, per-user-group model limits and usage visibility to control it. The practical move is to pilot agents with one team, measure consumption, and set group-level budgets before a company-wide rollout.
What should a small business do first?
Start with the basics that larger companies are also rushing to put in place: decide who may create agents, scope their data access, enable spending alerts, and document ownership. If you are weighing whether your processes are ready for this kind of automation at all, our guide to 5 signs your business is ready to automate with AI is a useful self-check before you commit.
Copilot's September 2026 overhaul is the clearest sign yet that AI at work is moving from answering questions to doing jobs. The businesses that benefit most will not be the ones that switch everything on first, they will be the ones that switch it on with identities, permissions, spending limits and ownership in place. If you want help designing that rollout, from tenant configuration to agent governance and cost controls, talk to the team at Skills Technology Solutions and we will map the new Copilot capabilities to your actual workflows.
Related Articles
Keep reading — more guides you may find useful.
Related Services
Services connected to this guide.
