Mott MacDonald says it is rolling Microsoft 365 Copilot out to all 20,000 employees and that staff have created more than 8,000 AI agents. The engineering firm describes AI already being used on live projects, including to flag conflicts in a set of 220,000 infrastructure requirements.
Microsoft AI Watch analysis
What happened
In a Microsoft UK Stories account, Mott MacDonald’s global head of AI and data, Nasrine Tomasi, describes a rollout that grew from 50 Copilot licences to 1,000 and then 5,000, before a new five-year agreement covering the full workforce. The company says it has also created more than 8,000 agents for work ranging from project assurance and stakeholder engagement to design automation.
One airport project used AI to examine about 220,000 requirements and flag around 200 conflicts for attention, the account says. On a separate underground-station project, Copilot-supported analysis helped designers show that temporary works could form part of the permanent structure, reducing additional construction.
Why it matters
These are concrete examples of AI moving into engineering workflows where documents, design choices and project coordination have real consequences. They also show the less glamorous work that comes with thousands of staff-built agents: Mott MacDonald says it has found overlapping tools, dormant agents and gaps in ownership and knowledge.
The firm is planning a central AI library to help staff find existing agents, and describes Microsoft Agent 365 as a future way to oversee the ecosystem. That oversight is still being developed, not a capability the account says is already in place.
Our read
The most useful part of this story is not the agent count. It is the reported work on live projects, including surfacing conflicts for engineers to examine. That is a better measure of practical adoption than a dashboard full of bots with names.
The scale is striking, but the figures and project examples come from Microsoft’s customer account, with Mott MacDonald’s AI lead describing the work. For engineering teams, the lesson is to build a way to find, review and govern agents as adoption grows, rather than assume a large collection will organise itself. It rarely does.
What to watch
- Whether the planned AI library helps staff reuse effective agents and retire redundant ones.
- How Microsoft Agent 365 will identify agent owners, data access and ongoing value.
- Whether Mott MacDonald reports outcomes from its proposal-writing agents, which aim to cut preparation time from a week to 48 hours.
- How engineers review AI-generated analysis before it informs project decisions.
Discussion spark: When engineers use AI to flag design or requirements conflicts, should the tool be judged mainly by how many issues it finds, or by how clearly people can verify its findings?
Sources and evidence
- Mott MacDonald: Engineering a responsible agentic AI future – Microsoft UK Stories (4 November 2026, 07:52 UTC)
not affiliated with or endorsed by Microsoft