Mott MacDonald staff have built more than 8,000 AI agents
Most enterprise AI rollouts start with a licence count and end with a usage dashboard. Mott MacDonald's numbers tell a different story about what happens when an engineering firm lets its people build. The consultancy began with 50 Copilot licences in 2025, moved to 1,000, then 5,000, and is now rolling Copilot out to all 20,000 employees under a new five-year Microsoft 365 E7 agreement. Along the way its people built more than 8,000 AI agents.
Who got access first
Half of the licences in the second wave went to early-career professionals, many of whom were already comfortable with generative AI. Nasrine Tomasi, Global Head of AI and Data, calls what followed "a little bit of reverse mentoring": younger staff learning engineering from experienced colleagues while showing those same colleagues what the tools could do. Many of them have since become AI champions inside the business. The firm took a lesson from that about access itself. Rather than trying to predict where AI would be most useful, it decided to give people the tools and let them apply their own expertise. "We've got very creative people. They love to think about how they can solve a problem," Tomasi said.
The work is specific. On one airport project, AI helped the team work through around 220,000 requirements and flag about 200 conflicts. A hydropower team built a set of agents to assemble requirements, calculate budgets and match capabilities for bids, refines them after every bid, and has set itself a target of cutting the average proposal process from a week to 48 hours. The internal assistant EMMA (Every Mott MacDonald Answer) has moved from finding information into monthly project reviews, and more than 10,000 lessons have been captured since that app launched in May.
What 8,000 agents cost
Bottom-up building has a bill attached, and the firm is candid about it. Rapid growth produced overlapping solutions, dormant agents and knowledge gaps when creators leave or change roles. The response is a central AI library, so a team that wants a proposal-writing agent can find out whether one already exists, plus a planned single view of who owns each agent, what data it can reach and whether it still delivers value. Tomasi describes that as a future "control tower".
What this means for L&D
The adoption programme is meant to help people use the tools confidently while still critically reviewing the output, so technical leaders can spot weak answers and know when human judgement takes over. That is the part most rollouts skip. Access creates builders. Builders create sprawl. Somebody has to teach the habits of ownership, review and reuse before the 8,001st agent, and that is the brief behind our AI Training programs.
Sources:
News and insights for innovation, digital transformation, future of work and L&D leaders.
Stay ahead of learning and development, corporate innovation and digital transformation news. Plus the future of work. For leaders in AU, NZ, HK, SG, the US, the UK and Canada.
