The 30-second version
- On July 7, 2026, Microsoft made Sales Agent and Service Agent generally available in Microsoft 365 Copilot, Dynamics 365, Outlook and Teams; two AI agents that work inside the tools sales and support teams already use.
- Both agents are powered by "Work IQ" and wired directly into live Dynamics 365 CRM data through the MCP (Model Context Protocol) standard; Service Agent alone adds more than 70 new MCP tools.
- Microsoft cites market figures to back the move: sales teams equipped with AI recommendations are said to be 2.6x more likely to hit their growth targets (Gartner), and agentic AI is expected to resolve 80% of common customer service issues on its own by 2029.
- The backdrop is a shift in the business model: Gartner estimates that $234bn of enterprise software spend is exposed to "agentic arbitrage" by 2030, which undermines per-seat pricing.
Microsoft has just moved two AI agents from demo to a product shipped to its entire CRM base. On July 7, 2026, the company announced general availability of Sales Agent and Service Agent, two assistants that act directly inside Outlook, Teams and Dynamics 365. The announcement looks technical; it actually reaches the core of how companies buy, and pay for, their business software.
What Microsoft announced on July 7
On July 7, 2026, Deva Rajamohan, Corporate Vice President of Dynamics 365 Customer Experience, announced the general availability of Sales Agent and Service Agent (CX Today). The two agents are not a separate app: they run inside Microsoft 365 Copilot, Dynamics 365 Sales, Dynamics 365 Customer Service, Outlook and Teams. The guiding idea is to remove the back-and-forth between the CRM and everyday tools.
Sales Agent produces account summaries, brings context to an opportunity and captures meeting follow-up in natural language. After a conversation, a seller dictates their key points, objections and commitments; the agent writes them straight into CRM fields without switching applications. Service Agent, for its part, summarizes a case, recommends the next action, drafts a resolution email and updates the record. At general availability it moves, in Microsoft's words, "from answering and summarizing to taking action across the entire service workflow", with more than 70 new MCP tools and around 20 core product enhancements (Microsoft).
The key technical point: both agents are powered by "Work IQ" and wired directly into Dynamics 365 data through MCP (Model Context Protocol), the open standard that lets an agent query a third-party system. Microsoft names two early customers, manufacturer Sandvik Coromant and bank Northern Trust, and leans on market figures to justify the investment: per Gartner, sales organizations equipped with AI next best actions are 2.6x more likely to achieve commercial growth, and agentic AI is expected to autonomously resolve 80% of common customer service issues by 2029, cutting costs by roughly 30%.
Why putting the agent in the flow of work changes things
Because the CRM's historic weak spot was never the feature set, it was data entry. A seller spends a meaningful share of the week re-typing into Salesforce or Dynamics what they just said on the phone; and when they skip it, CRM data decays, forecasts go wrong and the tool loses its value. By placing the agent where the work already happens, in email and the meeting, Microsoft attacks the problem from the right end: the seller no longer goes to the CRM, the CRM comes to record what was said.
That shift has a strategic consequence. As long as the agent lived in a separate interface, it stayed a tool you open "when you remember to". Embedded in Outlook and Teams, it becomes a quiet layer that runs continuously. The move to general availability on July 7 signals that Microsoft judges this layer reliable enough to bill across its entire Dynamics base, rather than keep it in a pilot program.
The real stakes: per-seat pricing wobbles
Behind the product launch sits a business-model question that Gartner put a number on six days earlier. On July 1, 2026, the firm estimated that $234bn of enterprise software spend is exposed to what it calls "agentic arbitrage" by 2030, roughly 20% of enterprise application SaaS spend (Gartner; CIO). The logic is simple: when one agent does the work ten users used to do by hand, the company no longer needs ten licenses.
George Brocklehurst, managing vice president at Gartner, sums up the reversal: "You are no longer buying software primarily for people; you are increasingly buying it for agents." Per-seat pricing, dominant for twenty years, assumes exactly the opposite: a fee per human user. If the agent absorbs the work of several people, the number of billable seats shrinks. Some vendors are already hedging by selling consumption units; ServiceNow, for one, offers SKUs billed per automation executed rather than per user.
Microsoft's position in this fight is a particular one. By embedding its agents into the Microsoft 365 and Dynamics 365 base it already sells, the company protects its footprint: instead of losing seats, it adds an agentic layer on top of the existing subscription. It is a way to stay on the right side of the arbitrage. Gartner's advice to buyers is worth keeping here: "Scrutinize the contract as much as you scrutinize the technology." The real negotiation over the coming months will not be about what the agents can do, but about how they are billed.
MCP, the plumbing that makes the agent useful
A sales agent is only worth anything if it sees up-to-date data. That is the job of MCP (Model Context Protocol), the open standard that lets an agent query and update a third-party system in a structured way. Microsoft wires Sales Agent and Service Agent into Dynamics 365, but also into Dataverse, SharePoint and Microsoft Graph, while respecting each user's existing permissions. In practice, the agent sees only what the person is allowed to see; a governance detail that matters a great deal to an IT department hesitant to let an agent loose on its customer data.
The choice confirms a deeper trend: MCP is becoming the standard connector between agents and business systems. The 70 MCP tools shipped with Service Agent are not a marketing line; they are that many access points through which the agent reads and writes in the CRM. The quality of a CRM agent will be judged less and less by its language model, and more and more by the cleanliness of the data and connectors that feed it.
What I take away for my own projects
This CRM data-entry story is one I live on every engagement. On the platform I built for e-Enfance on Salesforce, a large part of the work was integrating 3CX telephony so that conversations flowed automatically into the records, precisely to spare the teams the double entry that Sales Agent aims to remove. On my Horus engagement, a bilingual Pipedrive setup, same principle: a CRM is only as good as the reliability of what you put in it. Microsoft's announcement does not change that; it industrializes it. Still, plugging an agent into a badly structured CRM only automates the mess faster. Before dreaming of autonomous agents, the real value on the client side often sits upstream: cleaning the data schema, framing permissions and hardening the connectors. It is the less glamorous work, but it is what decides whether a sales agent keeps its promises or fills the CRM with noise.
The general availability of Sales Agent and Service Agent is not just one more product release. It marks the CRM agent leaving the lab for the invoice; and the real question for a leadership team is no longer whether these agents work, but how many seats they will make disappear.
Frequently asked questions
What do Microsoft Sales Agent and Service Agent do?
Sales Agent and Service Agent are two Microsoft AI agents, generally available since July 7, 2026 inside Microsoft 365 Copilot, Dynamics 365, Outlook and Teams. Sales Agent summarizes accounts and logs meeting follow-ups straight into the CRM; Service Agent summarizes cases, recommends the next action and drafts resolution emails.
What is agentic arbitrage, according to Gartner?
Agentic arbitrage describes a single AI agent replacing the work of several software users, which cuts the number of licenses a company needs. In a July 1, 2026 analysis, Gartner estimates that $234bn of enterprise application software spend is exposed to it by 2030, roughly 20% of the enterprise application SaaS market.
Why do these agents threaten per-seat pricing?
Per-seat pricing charges a fee per human user. When an AI agent absorbs the work of several people, a company needs fewer licenses and its software bill falls. That mechanism pushes vendors from per-seat toward usage-based or outcome-based pricing, such as SKUs billed per automation executed.
What does the MCP protocol bring to a CRM agent?
MCP (Model Context Protocol) is an open standard that lets an agent query and update a third-party system in a structured way. For a CRM agent, it provides access to live data in Dynamics 365, Dataverse or SharePoint while respecting each user's permissions. Service Agent ships with more than 70 MCP tools.