Two senior directors at a 5,000-person consultancy spent a bank holiday weekend vibe-coding an alternative to the Microsoft Dynamics CRM system their business had spent over a year and several million dollars building. By the following week, the project was on hold. Nicola Ray from Modern was on the call. "You had literally the CIO, chief sales officer, chief operations officer on a call going, we're going to put a halt to the project while we investigate this even further," she says.
It's one of several stories Nicola tells about AI adoption moving faster than the structures meant to contain it. At one client, a departing employee had built a personal Claude agent to generate monthly reporting graphs. When he left, his replacement typed in the same prompt and got a different result. "We didn't have access to his personal Claude, even though we've got Claude ourselves... It's his Claude and this is this person's Claude and it's trained in completely different way," Nicola says.
Budgeting for AI is its own problem. Nicola built a landing page using Lovable and found each prompt cost something different, with no way to predict the total. "How much does it cost to create a landing page? I don't know," she says. One of her clients, working in a Salesforce sandbox, moved a large volume of data somewhere it didn't need to be and burned through 15 million credits by accident.
Not every use case worries her. Asked about mining CRM call notes to track account-based marketing signals before results show up in quarterly reporting, she calls it "a really practical use of AI" for programmes with long sales cycles.
Modern has spent three years moving away from campaign execution and into pure strategy work, which Nicola credits for her confidence. "I know where I want to go is what they come to us with," she says. "I don't know how to get there and we help them with that how."
Listen to the full conversation with Nicola Ray on Aperture: Agency AI.
