What I advise on
Microsoft ships AI into Dynamics 365 at release-wave pace: Copilot inside the apps, and agents such as the Scheduling Operations Agent in D365 Field Service. Some of it is ready for daily operations, some of it is preview that will change under your feet. The advisory question is never whether AI is coming; it is what deserves your attention now, what can wait a wave, and what your data can actually support.
- A readiness assessment of AI and Copilot capabilities for your D365 environment
- Pilot design: sandbox, your own data, measurable criteria, your most skeptical users
- A clear go, wait or no per capability, with the preview-versus-production status spelled out
- Data readiness: an agent can only be as good as what the system knows
Pilots, not promises
The most expensive AI mistake I see is architecting around a preview feature as if it were finished. The second most expensive is ignoring the direction entirely. The way between those two is a contained pilot: a sandbox, a real week of your own work, and criteria agreed before it starts. I published my current read on scheduling agents in Do you still need RSO?, which is exactly the kind of assessment this service produces for your situation.
Why an independent voice helps
Every vendor demo shows the best case. Someone who implements Dynamics 365 every day knows the difference between what demos well and what runs well, and has no license revenue riding on your answer. If a partner selection is part of the same decision, that is its own service.
AI in Dynamics 365 questions
Which AI capabilities in Dynamics 365 are ready for production today?
That answer changes per release wave, which is precisely why the assessment exists. As a rule of thumb in 2026: several Copilot capabilities inside the apps are generally available, while agent capabilities such as the Scheduling Operations Agent in D365 Field Service are still in preview. I keep the status per capability current and spell it out in the advice.
What does a pilot look like?
A sandbox environment, a real slice of your own data and work, measurable criteria agreed upfront, and your most skeptical users at the controls. At the end you have evidence instead of a demo impression.
Is our data ready for AI?
That is usually the real question. Agents and optimizers read what the system knows: skills, assets, preferences, history. If that lives in spreadsheets next to the system, the pilot will tell you quickly, and fixing the data flow becomes step one of the roadmap.