The Iceberg Principle: Why Spud Looks Simple and Why That's the Whole Point
By Polly Barnfield, OBE, CEO of Maybe*
Behind one Microsoft Teams chat sits an orchestration layer connecting Float, Box, brand compliance, news, and synthetic audiences. The team only sees Spud. That is the entire design philosophy in one sentence.
The first and most important thing Maybe* and Liganova MaSH! built together was not an agent. It was an interface.
Everything flows through Microsoft Teams. Team members talk to one agent, called Spud. They do not need to know which model is running under the hood, which agent has been tasked with the job, or which system is being accessed. They just ask. The work gets done.
Spud is not a chatbot. Spud is an orchestration layer. Behind a single Teams interface sits a set of agents connecting to Float for resource planning, Box for file management, a brand compliance checker that knows BMW's CI inside out, a news aggregation engine, and a synthetic audience capability unlike anything else in the agency's stack. The complexity is invisible. The interface is the one the team already uses every day.
“It feels really small, but technically it’s an iceberg behind the scenes because you’ve built a monster to deliver a very thin surface that just makes people think, wow, that’s made my day better. And that’s the end result.”
This is the architectural choice that defines this edition. And it is the same choice every client we will ever profile in Scaling Up With Agents has to face. Do you make the team learn your AI, or does your AI live where the team already works? The wrong answer is cheaper to build. The right answer is the only one that produces actual adoption.
What sits behind the Teams chat
Without giving away the engineering, Spud orchestrates five distinct capabilities. Each one started life as a problem surfaced in those sixteen interviews.
Brand compliance checking. Designers were spending significant time on CI checking before work went out the door. The brand policy requirement meant multiple people needed to review everything. An agent now checks work against their client’s brand guidelines in real time as it is created, reducing the review burden and getting work out faster. Crucially, humans are still in the loop. They are checking the agent's output rather than doing the grunt work themselves.
Resource planning. When a brief came in or shifted, a round of phone calls and Teams messages would begin. Can I borrow this designer? Is that person available? Do we have the right skill set? Float is now connected to Teams. Availability is checked, and the time is booked, in a single message. Three phone calls become one.
File finding. During discovery it became clear that searching for files across a growing archive was taking far more time than it should. Multiply that across a whole agency and the loss compounds daily. Teams now surfaces files from Box on demand. Natural language search, not folder navigation.
News on demand. Any team member can have relevant news delivered from across the web directly into Teams, sliced by topic, client, or brief. No browser tabs. No aggregator tools. The information arrives.
Synthetic audiences. This is the capability James describes as the one he is most excited about, and it deserves its own paragraph.
The synthetic audience piece
This is not just an efficiency tool. It is changing what the agency can offer.
The capability lets the team take online and social audiences and replicate them synthetically, so creative work can be tested, validated, and optimised before it goes live. Clients are already using it for product development before expensive tooling begins. The agency uses it for pitch work, brief stress-testing, and audience segmentation into specific niches.
“Whether it be a proactive inspiration piece, clients using it for product development, stress testing creative ideas, or segmenting content into specific niche audiences, which is really important to be relevant in our world, the synthetic audience is the piece I’m most excited about.”
This is what an agentic capability does that a workflow cannot. It does not just save time on existing work. It opens up work the agency could not previously offer at all.
The naming question, which matters more than it sounds
Calling the agent Spud was not a small decision. It connects directly to MaSH!. It is approachable. And it made something that could have felt like surveillance or replacement feel instead like a colleague.
“By making it approachable and humanising it is really important. It stops it being a piece of technology and makes it an extension of their team.”
The creatives in the agency had the most scepticism going in. They were worried that the agency was about to ask a machine to do their job. The naming, the approach, and the visible focus on removing admin rather than replacing creativity were what brought them along.
“Not jumping to a Gen AI solution probably helped us with the most concern in terms of are you going to just ask it to do all the creativity. We showed them we were going on this journey, looking at both efficiency and effectiveness. And those champions that were jumping in two feet first are slowly bringing the enthusiasm right throughout the agency.”
The principle this is teaching the wider market
If you take one thing from this post, take this. Most failed AI deployments fail not because the technology is wrong, but because the surface is wrong. The team is being asked to log into a new tool, learn a new chat interface, switch context constantly, and remember which task goes to which AI. That friction is where adoption goes to die.
The Liganova MaSH! deployment works because the AI lives inside the surface the team already uses. Microsoft Teams. One chat. One name. The infrastructure behind it can be as sophisticated as the problem requires. The team's day stays simple.
“If you make the AI layer become integral to where you already work, the team just have colleagues that help make them work better. Don’t make it another system they have to log into and learn how to use. It has to be part of the surface they already use.”
Next post is the worked example. We will follow a single creative brief from arrival in the inbox to client delivery, and show what each Spud capability actually does at each step. The point is to show the iceberg in motion.
The full Spud story is on the FutureWeek podcast. Listen to the conversation here.
If your team is logging into too many AI tools and adoption is suffering, book a call with us.
