The short version
Copilot Cowork can now work inside Dynamics 365 finance and operations through three tool types: Data tools that read records, Form tools that complete fields on screen, and Action tools that run a process. Microsoft's Work IQ orchestrates the steps. It runs under the signed-in user's security roles and presents any write for human review before it commits. As of mid-2026 the ERP capability is in preview through the Frontier programme, so treat it as something to pilot in a sandbox, not to depend on in your close.
What Copilot Cowork Is
Copilot Cowork is Microsoft's agentic layer for Copilot: instead of only answering questions, it can carry out multi-step work across an application. The wider Cowork capability went generally available earlier in 2026. What is newer, and the reason it matters to finance, is that Microsoft has been extending Cowork to act inside Dynamics 365 finance and operations, the system where the ledger, sub-ledgers and procurement actually live.
The difference from a normal Copilot prompt is that Cowork can string together the steps a person would otherwise click through: find the relevant records, open the right form, populate the fields, and run the process. The orchestration is handled by Microsoft's Work IQ, which sequences the tools and keeps track of context across the steps.
The ERP Capability Is a Preview
This is the most important caveat, so it goes near the top rather than buried at the end. As of mid-2026, the capability that lets Cowork operate inside Dynamics 365 finance and operations is in preview, released through Microsoft's Frontier early-access programme. It is not a general-availability feature you should assume is stable, complete, or contractually supported for production finance work.
In practice the preview has carried prerequisites that move over time: a recent finance and operations version (version 10.0.45 or later has been called for), the relevant Copilot and Frontier entitlements, and a non-production environment. It has also not supported Cloud Hosted Environments. Because all of this changes between releases, confirm the current status in the Microsoft 365 admin centre and the Dynamics 365 release plans before you rely on anything here. This is a living topic, and we update the page as the status moves.
A Sheffield engineering firm of around 300 staff shows why this caveat matters in practice. Its board had seen a Cowork demo and wanted it live in the finance team within the quarter. On checking, the firm was running an older finance and operations build on a Cloud Hosted Environment, which the preview does not support, so the realistic answer was a sandbox pilot once they upgraded, not a production rollout. Knowing the prerequisites before promising the board saved an awkward conversation later, and it is the first thing we check on any Cowork enquiry.
Want help scoping a safe Cowork pilot for your finance environment?
Talk to us about a Dynamics 365 AI pilotData, Form and Action Tools
Cowork acts inside Dynamics 365 through three distinct tool types, and the distinction matters for understanding the risk.
Data tools. These read from the system: pulling records, looking up balances, gathering the figures behind a question. Read-only by nature, this is the lowest-risk category and where most early value sits, because so much finance work is really information retrieval across forms.
Form tools. These complete fields on a form the way a user would, putting proposed values into the right places on screen. The values are visible and editable before anything is committed.
Action tools. These run a process, the equivalent of clicking the button that posts, confirms or submits. This is the category that writes to the system, and it is the one held behind a review step.
Security Roles and Review Before Write
Two design choices make this defensible for finance. First, Cowork runs under the signed-in user's existing security roles. It cannot read or do anything the person directing it could not already do manually. Your existing segregation-of-duties and role design carry straight over; there is no separate super-user identity to govern.
Second, anything that writes is presented for the user to review and approve before it commits. Cowork proposes the form values or the action; the finance person checks and confirms. That keeps the human firmly on the control point that matters, which is the moment something changes in the ledger. The right mental model is an assistant that does the clicking and gathering, with the qualified person still making the decision and owning the sign-off, exactly the governance model we describe in our AI in finance guide.
Where It Helps Finance Teams
The early wins are the high-friction, multi-form tasks that eat time without needing judgement: retrieving and assembling information that currently sits across several screens, preparing a form for review, and walking a standard process to the point of approval. A finance analyst spending part of every day navigating between inquiry forms to answer routine questions is the clearest candidate.
It sits alongside, rather than replacing, the other AI layers a finance team uses. Copilot in Excel and Outlook handles the document and email work, Claude handles the heavier reasoning and narrative, and Cowork inside Dynamics 365 handles the work that has to happen in the system of record. Our guide to AI in ERP across SAP, Dynamics and Workday puts this in the wider context, and our Microsoft Copilot for business page tracks the broader Copilot picture.
Free 30-minute consultation
Map your AI opportunities
Book a free 30-minute consultation. We’ll review your workflows, identify the highest-ROI AI opportunities, and tell you whether you need consulting, training, or a fractional CAIO. No pitch, just direction.
Book a 30-minute callHow to Approach a Pilot
Because the ERP capability is in preview, treat this as a controlled experiment rather than a rollout. Pick one bounded, repetitive workflow with a clear owner. Stand it up in a sandbox, confirm the security roles are exactly right for the people who will use it, and define who reviews and approves the writes. Run it against realistic but non-live data, measure the time saved against the manual baseline, and only then consider whether the status and your testing justify any production use.
If you want help scoping that safely, this is the kind of work our AI consulting and Dynamics 365 engagements do. An AI readiness assessment identifies which finance workflows are the right first candidates, and our AI for finance training gets the team using these tools with the right controls.