On July 7, 2026, Microsoft announced that its Sales Agent and Service Agent in Microsoft 365 Copilot are generally available. Not a preview. Not a waitlist. General availability means eligible Microsoft customers can now deploy them outside a preview program. Access still depends on the right Microsoft 365 Copilot and Dynamics 365 licenses, a connected CRM, administrator setup, and user permissions.
That changes the question every leader has been asking.
For two years the question was some version of "should we experiment with AI agents?" You could study it, pilot it, or wait. Now the largest software vendor most companies already write a check to has answered part of that question for you. The agents are here. The question that falls to you is different, and it is harder: who decides how they get used, and where does a person stay accountable for the results?
What actually went generally available
The July 7 post from Deva Rajamohan, Corporate Vice President for Dynamics 365 Customer Experience, covered more than two agents. Here is the plain list.
- Sales Agent in Microsoft 365 Copilot is now generally available.
- Service Agent in Microsoft 365 Copilot is now generally available.
- Microsoft 365 Copilot in Dynamics 365 Sales and Microsoft 365 Copilot in Dynamics 365 Customer Service are both generally available.
- A Dynamics 365 Sales plugin for Copilot Cowork and a Dynamics 365 Customer Service plugin for Copilot Cowork round out the lineup.
Two customers spoke in the announcement. Silvana Zafarana of Sandvik Coromant said, "Sales Agent represents an important step in our broader agentic journey." DP Indetkar of Northern Trust pointed at how service work is changing: "What excites us about Service Agent is the move from reactive search to proactive intelligence."
Read those two quotes again. Both customers point less at the features and more at the work changing shape underneath the people who do it. That is the part worth your attention.
Why "generally available" is a different kind of pressure
A preview is a limited program. You opt in, you assign a small team, you learn. General availability changes what comes next. For an eligible organization, the capability can move from a small pilot into broad deployment. That puts the pressure on leaders to decide who gets access, what the agent may do, and where a person reviews the work.
The vendors have numbers that explain why they are pushing. A Microsoft-sponsored IDC study, "2024 Business Opportunity of AI," surveyed more than 4,000 business leaders and AI decision-makers worldwide and reported an average return of $3.70 for every $1 invested in generative AI, with top performers seeing $10.30. Treat that as a vendor-commissioned figure, because it is one. It tells you the ceiling some companies reach. It does not tell you that you will reach it.
The independent research points somewhere more useful. Gartner predicts that agentic AI will autonomously resolve 80% of common customer service issues by 2029, cutting operational costs by 30%. On the sales side, a Gartner survey of 227 chief sales officers, conducted from August through September 2025, found that sales organizations providing AI-enabled next best actions are 2.6 times more likely to achieve commercial growth.
Gartner found sales organizations providing AI-enabled next best actions are 2.6 times more likely to achieve commercial growth, based on a survey of 227 chief sales officers conducted August through September 2025.
Source: Gartner, 2026Look closely at that sales finding. Gartner tied the stronger results to AI-enabled next best actions and to organizations that redesigned their seller workflows. The survey shows an association. It does not prove that the AI or the redesign caused the growth. Even so, the pattern is worth reading. As these agents reach more properly licensed organizations, access alone becomes a thinner advantage. Next best actions are not a feature you switch on. They are a decision about which judgment stays human and which steps the agent can carry.
The differentiator is proficiency, not access
Here is what trips up smart leaders. When a capability is available to everyone, access stops being the advantage. If your competitor down the street can turn on the same Sales Agent by lunchtime, the agent itself is not what separates you.
What separates you is whether your people can run it well.
That is what The 7 Levels of AI Proficiency measures. It is a way to see, honestly, where a person or a team sits on the climb from first curiosity to designing the workflows that agents run inside. A team at the early levels will treat Sales Agent like a faster autocomplete and get faster mediocrity. A team further up the climb will ask the harder questions. Which parts of our sales process actually benefit from an agent taking the next action? Where does a human need to check the work before it reaches a customer? What does "good" look like, and how will we know if the agent drifts from it?
The Gartner 2.6x finding is proficiency wearing a business suit. The organizations pulling ahead built the muscle to redesign the work, not just to adopt the tool.
What this means for the decisions you own
You do not need to become technical to lead this well. You need to make a few calls that only you can make.
Decide where a human stays in the loop. Gartner's 80% figure is a forecast about the market by 2029. It is not a measured result, and it is not a promise about Microsoft's Service Agent. Even if it proves directionally right, autonomous resolution of common issues still needs someone accountable for what the agent does with the uncommon ones, for how it treats a customer on a bad day, and for what happens when it is confidently wrong. Name that person. Name that boundary. Do it before the agent is live, not after a mistake reaches a customer.
Decide what you are measuring. Vendor ROI figures are a starting point, not a target you inherit. Pick the outcome your P&L depends on. Faster response times. Higher win rates. Lower cost per resolved ticket. Then hold the agent to it the way you would hold a new hire to it.
Decide who is building your team's proficiency. The agents are available now. The judgment to run them is not automatic, and it does not arrive with the license. Someone has to own the climb, whether that is a leader inside your company or a partner who does this for a living. This is the work behind the phrase "give AI an owner." Not the software. The results.
The next step
You do not have to decide the whole strategy this week. You do have to answer one question before the agents get used by default: for your first agent, in sales or in service, where does a person stay accountable for the outcome?
Write that answer down. Put a name next to it. That single decision is the difference between an AI agent that carries your team forward and one that quietly makes your problems move faster.
Related reading: Level 5: The Captain (Design Thinker).
Sources
- Moving sales and service organizations forward with agentic CX and Microsoft 365 Copilot
- Generative AI is delivering substantial ROI to businesses (Microsoft-sponsored IDC report)
- Gartner Predicts Agentic AI Will Autonomously Resolve 80% of Common Customer Service Issues by 2029
- Gartner Survey Finds Sales Organizations With AI-Enabled Next Best Actions Are 2.6x More Likely to Achieve Commercial Growth
Frequently Asked Questions
Are these agents fully autonomous?
Microsoft describes agentic capabilities across sales and service, and Northern Trust described Service Agent as a step toward proactive intelligence. The honest read is that autonomy is a spectrum, not a switch. Gartner's own prediction is that agents handle common issues by 2029, which leaves the harder cases, and the accountability for all of it, with your people.
Do we have access already?
Not automatically. General availability means the release is out of preview. It does not mean the agents are switched on for every Microsoft 365 user. Access depends on your Microsoft 365 Copilot and Dynamics 365 licensing, a connected CRM, user permissions, and administrator setup. Confirm the exact requirements with your Microsoft licensing contact before assuming your team has access.
What is the risk of moving too fast?
The risk is the process itself. Deploying an agent onto a sales or service process nobody redesigned first is where it goes wrong. An agent amplifies whatever workflow it inherits. Give it a broken process and it will run the broken process faster.
What is the risk of moving too slow?
The Gartner sales finding suggests the companies redesigning their work around AI-enabled next best actions are already pulling ahead. Waiting is a choice, and it has a cost, but it is a smaller cost than deploying without a plan for oversight.
Find your AI Proficiency level
The free 7 Levels assessment places you across seven stages of AI capability. Under ten minutes. Research-backed scoring.