Artificial intelligence is increasingly evolving from a supporting tool into an integral part of business operations. Employees are no longer using AI agents solely for individual tasks but are increasingly delegating multi-step processes, entire projects and recurring activities to them.
This trend is also reflected in the latest adoption figures: more than 30 million paid Microsoft 365 Copilot seats are now deployed across organisations worldwide. Compared with the previous quarter, net seat growth has doubled.
As a result, the benchmark for measuring enterprise AI success is changing. In addition to adoption, licences and time savings, the focus is shifting towards how AI transforms work, enables new capabilities and becomes specialised for specific roles and business processes.
This transformation is becoming evident in three key areas:
- Everyday work: AI is becoming an integral part of daily work. Employees are increasingly delegating tasks, projects and multi-step processes to AI agents. Copilot Cowork can independently plan, execute, review and, where necessary, correct such work. In testing, Cowork was 30 to 40 per cent more cost-effective than solutions based on other models.
- New possibilities: AI is changing what organisations are capable of achieving. While AI accelerates processes and improves efficiency, organisations are increasingly using it to fundamentally redesign the way work is done and unlock new opportunities. Microsoft’s Cloud Supply Chain, for example, first streamlined six end-to-end processes before deploying more than 70 specialised AI agents. In selected workflows, this reduced cycle times by 75 per cent.
- Specialised roles: AI is becoming tailored to specific roles. Success is no longer determined solely by how broadly organisations deploy AI, but by how precisely they adapt it to the actual work of individual roles. One example is Microsoft’s sales organisation, where different sales profiles have been paired with specialised AI agents. As a result, the share of time spent on customer-facing activities increased from 25 to 50 per cent. In the pilot group, revenue per salesperson increased by 9.4 per cent, while the time required to close deals was reduced by 20 per cent.
For further information and examples, please read the blog post by Jared Spataro, Chief Marketing Officer, AI at Work.
***
¹This analysis was conducted by internal Microsoft teams. It compared 125 test runs using a total of twelve simple, medium-complexity, and complex work-related data prompts in Copilot Cowork and Claude Cowork, each with the respective Microsoft 365 connector. Both systems used the Opus 4.8 model.
For Copilot Cowork, costs were calculated based on variable rates for the relevant models, context, tools used, and runtime, using internal log data. For Claude Cowork, costs were calculated based on publicly available API pricing, taking into account token usage and use of the Microsoft 365 connector.
On this basis, the total cost per test run was determined, and the costs of simple, medium-complexity, and complex prompt sets were compared between Copilot Cowork and Claude Cowork, each with the respective Microsoft 365 connector.
The cost-saving results are based on tests conducted in June 2026. Actual costs and potential savings may vary depending on usage, configuration, timing, and other factors.