Mott MacDonald AI Agents Top 8,000 as Copilot Reaches All Staff
Engineering consultancy Mott MacDonald says its employees have built more than 8,000 AI agents in Microsoft Copilot, and the firm is now giving Copilot to its entire workforce of roughly 20,000 people. Nasrine Tomasi, the firm's global head of AI and data, outlined the phased rollout in an interview with Microsoft, as reported by Construction Management on November 4, 2026.
- Agents: more than 8,000, all created by staff on Microsoft Copilot
- Licence growth: 50 seats at first, then 1,000, then 5,000, now expanding to all 20,000 employees
- Who got access: half of the second-wave licences went to early-career professionals
- Data layer: Microsoft Fabric is being readied as a central data platform for Copilot and its agents
- Governance: a central AI library is under development to reduce duplication
- EMMA: the in-house assistant now runs monthly project reviews in Microsoft Teams, logging over 10,000 lessons since May
How Mott MacDonald uses its Copilot AI agents
The examples span design, assurance and bidding. On an airport job, an agent helped the team sift through about 220,000 requirements and surfaced around 200 conflicts for engineers to resolve.
On a large underground station, structural designers used Copilot-assisted analysis to demonstrate that temporary works could safely stay in place as part of the permanent structure. That reduced the extra construction the project needed.
A hydropower team has built a set of agents for proposals. They pull together client requirements, work out budgets and map opportunities to the firm's capabilities. The team tunes the agents after every bid and aims to shrink a typical proposal from a week to 48 hours. Tomasi credited the approach with helping the group win more work, and said the firm's focus has shifted from speed alone toward quality and client value.
Managing thousands of staff-built agents
Growth at this pace has created problems Mott MacDonald openly acknowledges: overlapping tools, agents nobody uses anymore, and lost know-how when an agent's creator changes roles or leaves.
The answer is a central AI library. Before building, say, a proposal-writing agent, staff will be able to check whether one already exists and reuse or adapt it. The firm also wants the library to reveal which agents deserve wider rollout.
EMMA moves from search into project reviews
EMMA, short for Every Mott MacDonald Answer, began as a knowledge-retrieval assistant. It now supports monthly project reviews through a Teams app that records a structured conversation between the project manager and project principal, then sorts the notes against a set agenda with lessons and next actions. That replaces a standard health-and-status form.
What is still unknown
The headline figure counts agents created, not agents in active use. Mott MacDonald has not said how many of the 8,000 are used regularly, how agent outputs are checked before they inform design decisions, or what the program costs.
It is also unclear whether any agents connect directly to design or BIM authoring tools, or work only with documents, data and Microsoft 365 content. The claim of winning more work has not been backed by published figures.
What it means for AEC teams
For firms weighing Copilot at scale, the lesson is less about the agent count and more about the sequence: a small pilot, staged expansion, early-career users as champions, and then a data platform and agent registry once sprawl appears.
Teams starting out should set up an agent catalog and named owners from day one, define who signs off agent-assisted engineering work, and confirm their project data is organized well enough for agents to use. Requirements checking and bid preparation look like the most transferable starting points.
Recent Articles
Learn about the latest architecture software, engineering automation tools, & construction technologies
