Dynamics 365 agents can now see beyond the CRM record, and that is a bigger upgrade than the release notes make it sound. On 16 June 2026 Microsoft made the Work IQ API generally available, and at the end of June it confirmed general availability for Sales Agent and Service Agent inside Microsoft 365 Copilot, Outlook, Teams and Dynamics 365. The practical effect is that an agent working an account no longer reasons only over what a salesperson remembered to type into a form. It can also reason over the meeting that happened last Thursday, the thread with procurement, the specification document sitting in SharePoint, and the case history in Customer Service, all inside the permissions the user already holds.
That is a genuine gain for organisations that have spent years being told their CRM data was not good enough for AI. A great deal of the context that makes a commercial judgement sound was never going to live in a CRM field. It lived in calendars, inboxes, shared files and conversations. Work IQ is the layer that makes that material legible to an agent without asking anyone to re-key it, and it arrives with governance built in rather than bolted on afterwards.
The capability is real, and the teams getting the most from it are the ones treating it as a chance to widen what their agents know rather than as another feature to switch on. Here is what has actually changed, and where the early returns are showing up.
What is Work IQ, and what does it change for Dynamics 365?
Work IQ is Microsoft’s workplace intelligence layer for AI agents. It gives an agent a permission-aware way to access and reason over organisational data across Microsoft 365, including mail, calendar, files, people, chat and sites, as well as connected external systems. It reached general availability on 16 June 2026, is billed through consumption-based Copilot Credits rather than a separate licence, and exposes itself through a REST API, an agent-to-agent interface and an MCP server. For Dynamics 365 the significance is that Sales Agent and Service Agent are grounded in live CRM data and in Microsoft 365 work signals at the same time, so a single question can be answered from both without a custom integration project sitting underneath it.
Anyone who has scoped that kind of unified context before will recognise how much work has just been removed. The old route involved connectors, a warehouse, a semantic model, a refresh schedule and a long argument about which system was authoritative. Work IQ collapses a large part of that into a platform capability. Microsoft’s own developer material describes the MCP surface as roughly ten generic tools covering fetch, create and update operations, with a schema discovery tool so agents can learn structures at runtime instead of having every field hard-coded at build time.
Service Agent shows what that enables in practice. At general availability Microsoft shipped more than seventy new MCP tools alongside twenty-plus product enhancements, letting the agent draw on case records, knowledge in Dataverse and SharePoint, customer timelines and related service activity to produce cited answers inside the flow of work.
What is CRM data quality when agents can read your whole workplace?
CRM data quality is the degree to which the records in your CRM are accurate, complete, consistent, timely and unique enough to support the decisions people and systems make from them. The traditional definition focuses on fields: is the account name right, is the close date current, is the contact still employed there. That definition still matters, and it now sits inside a wider one. When an agent can also read the meeting invite, the thread and the document, quality becomes a question of whether the signals across your whole workplace point in the same direction, and whether the CRM remains the place where a decision is finally recorded.
This is unusually good news for organisations part-way through a data programme. You no longer need perfect CRM hygiene before an agent can be useful, because the agent can corroborate a thin record against the surrounding evidence. A close date that has not moved in two quarters reads differently when the agent can also see that the last customer meeting was six weeks ago and the procurement thread went quiet. That is a more honest picture than either source gives alone, and it is available immediately rather than at the end of a cleanup project.
The teams gaining most from this are being deliberate about one thing: they still write the conclusion back to the CRM. Context can come from anywhere, but the decision, the commitment and the number belong in a governed record that a forecast can be built on. Keeping that discipline is what turns a wider evidence base into better reporting rather than into a second, informal system of record.
How does permission-aware AI protect sensitive CRM and workplace data?
Permission-aware AI means every agent request runs in the context of a specific user and returns only what that user is already allowed to see and do. Work IQ implements this with a two-tier model: broad permissions establish the outer boundary, and a policy engine evaluates each individual request against resource paths, request methods, user identity and data content. An agent asked to summarise an account for a sales manager cannot surface a document that manager could not open directly, and the same question from two people can legitimately return two different answers.
This is the design decision that makes the wider context usable at all. Reading across mail, meetings and files sounds alarming until you recognise that the access model is the one your organisation already runs and already audits. Nothing new is being granted. The agent is simply a very fast reader operating inside existing boundaries, which is a far easier proposition to take to a security review than a pipeline that copies everything into a separate store first.
It does put a premium on permissions hygiene, and that is worth treating as an opportunity rather than a chore. Organisations that tidy over-broad SharePoint sites and stale Teams membership before rollout tend to find the exercise pays for itself twice, once in agent answer quality and once in a compliance posture that was overdue for attention anyway. It is a well understood piece of work with a clear finish line.
How do you improve CRM adoption when the agent sits in Outlook and Teams?
You improve CRM adoption by reducing the distance between where people work and where the record lives, then making the recorded outcome visibly useful to the person who created it. The Copilot Cowork plugins for Dynamics 365 Sales and Customer Service, generally available since June 2026, take a direct run at the first half of that. Account preparation, case triage, drafting and approved CRM updates now happen inside Microsoft 365, so a seller updating an opportunity does not have to break context to do it. The second half is still yours to design: people keep records current when the system gives something back.
This is where the adoption story genuinely improves rather than merely shifting location. For a decade the honest answer to low CRM adoption was that data entry cost the seller time and returned value to somebody else. When the agent drafts the update from the meeting that just happened and asks for confirmation, that trade changes. The seller spends seconds approving rather than minutes typing, and gets a better-prepared next meeting in return.
The word doing the work in Microsoft’s own description is approved. Governed write-back, where a human confirms what enters the record, is the pattern that lets you move quickly without loosening control. Decide early which fields an agent may propose, which require confirmation, and which stay manual, and you can expand that list confidently as trust builds.
Agentforce or Microsoft Copilot: how should a CRM buyer read this?
Both Salesforce Agentforce and Microsoft’s Copilot and Work IQ combination now offer agents that act on CRM data rather than only summarising it, and both bill through consumption rather than seats alone. The clearest difference is the centre of gravity. Agentforce is strongest when the work already happens inside Salesforce and its surrounding data platform. Microsoft’s approach is strongest when the work happens across Microsoft 365 and Dynamics 365 together, because Work IQ treats the everyday productivity estate as first-class context. For most buyers the honest selection question is not which vendor has better agents, but where their people actually spend the day.
That framing is liberating rather than limiting. Organisations already standardised on Microsoft 365 have a shorter path to value here than they have had with any previous wave of CRM AI, because the context layer is already populated and the identity model is already in place. Organisations running a mixed estate have a real choice to make, and they can make it on evidence: run the same three commercial questions through each platform against your own data and compare the answers.
Consumption pricing deserves the same practical treatment. Copilot Credits make cost a function of usage, which is genuinely fairer than paying for licences that go unused, and it rewards teams that measure. Pick a handful of high-value scenarios, watch the credit consumption in the admin centre for a few weeks, and you will have a defensible unit economic before you scale. That is a far better position than negotiating a large commitment on a forecast nobody can substantiate.
Why use an independent CRM partner for an agent rollout?
An independent CRM partner is a consultancy that implements and advises across multiple CRM platforms without being tied to a single vendor’s licensing incentives. On an agent rollout that independence is practical rather than philosophical. Someone has to judge whether a given process belongs to a Dynamics 365 agent, a Copilot Cowork plugin, a Power Platform flow or a person, and that judgement is cleaner when it is not shaped by which product line the adviser is measured on. The same independence helps when the answer is that an existing capability already covers the need.
The strongest programmes we see start narrow and deliberately. One process, one clear measure, one governed write-back path, then expansion once the pattern proves itself. That approach suits this release particularly well, because the platform work has already been done by Microsoft. What remains is the design work: which context matters, who confirms what, and how success is measured in commercial terms rather than in usage counts.
The Sirocco perspective
We think this is one of the more useful things to happen to Dynamics 365 in a while, precisely because it reduces the amount of preparation an organisation needs before agents earn their keep. Our teams across Stockholm, Barcelona, Dubai and Turin have spent years helping clients close the gap between what the CRM knows and what the business knows. Work IQ narrows that gap at the platform level, which frees our conversations with clients to focus on the parts that genuinely need judgement: process design, governed write-back, permissions readiness and the measurement that proves value.
Our advice is to move now and move small. Pick one commercial process where better context would obviously help, check that the underlying Microsoft 365 permissions reflect how the organisation actually works, define which updates an agent may propose and who confirms them, and measure the result against something a finance director would recognise. Because Sirocco works across Salesforce, HubSpot and Dynamics 365, we can tell you honestly when this is the right route for your estate and when a different one would serve you better. Either way, the opportunity in front of Dynamics 365 customers this quarter is unusually well timed, and the organisations that start now will be several cycles ahead by the time this becomes standard practice.
If you are weighing up how Work IQ and the new Dynamics 365 agents fit your own CRM estate, we would be glad to talk it through. Schedule a consultation with our team.
Get in Touch
Tell us which Dynamics 365 process you would most like an agent to support, and we will help you work out what context, permissions and approval steps it needs to be trusted with it.
