The old team shape stopped fitting. Here is how Removery rebuilt around AI agents.

When Michael DeHaven started as Removery’s Chief Systems and Digital Officer, he told his CEO he would “look like a nut for a couple months” because he didn’t know what his team should look like. Today, one designer, one product manager, and one engineer direct a team of AI agents. It replaced a model of about four engineers for every designer and product manager.

Summary: At Birdeye View 2026, Michael DeHaven, Chief Systems and Digital Officer at Removery (165+ studios), told Birdeye’s Chief Customer Officer Sanjay Kini how he rebuilt a technology team around AI agents. He moved to pods of one designer, one product manager, and one engineer, hired for experience, and built the data foundation agents need. This post covers each choice. Our companion post covers how Removery verifies AI output and keeps customer care human.

Why did Removery rebuild instead of patch?

DeHaven inherited several codebases, thin documentation, and a legacy approach from a former engineering team. He read through code himself, then sent AI agents across every project to assess what he had. The decision was clear: keep the lights on and build new.

He told the executive team what that meant. Freeze everything except vital fixes, then move fast on a new platform. That also meant a hard decision: move to a completely new remote team, and for a few weeks, he acted as the company’s lead engineer. The reason is Removery’s own shape. Its studios are lean, and an average one has two or three people busy with clients, so the technology behind them has to do more.

The pace since has been high. DeHaven said the team has pushed code more than 150 times, and that he has made more GitHub commits in five months than in the previous 30 years.

What does a pod look like with AI agents on the team?

DeHaven used to work with a ratio of one designer to one product manager to about four engineers, with QA supporting the engineers. Now it’s one-to-one-to-one: a designer, a product manager, and a full-stack engineer, with QA as a shared pool.

Each engineer directs a team of agents. His engineers tell him they run 1 to 4 work streams at once, depending on complexity. He said it’s “more like being the conductor of an orchestra” than a lone developer. It is not free, since running that many agents burns a lot of tokens, but the efficiency is “really interesting.” What the team does now, he said, is completely different from what it did 18 months ago.

He still calls this “experimental territory,” and he is open about the cost. His boss warned him that working at this pace can lead to burnout, and DeHaven agreed the balance was off. “Just because I can do what I did last night doesn’t mean that I should,” he said.

What did the pod do outside engineering?

DeHaven has led marketing, engineering, and product teams over his career, so on one project he wore all three hats. While marketing went through staffing changes, he took on a new landing page himself. He used Claude Code, design agents, and his experience to add Mixpanel tracking and build a new booking flow with calendar management and database integrations. Split tests were running in about four weeks.

By his account, the new booking experience converts to leads at about five times the rate of the previous flow, and Removery is shifting more traffic to it. He said the infrastructure built underneath is “more of a launchpad” for what comes next.

Why did DeHaven hire the more experienced developer?

He interviewed two candidates with equally good personality fit. When he pressed on how each used AI, one trusted the tools a lot, and the other was very skeptical. The result was counterintuitive: the more experienced developer trusted the tools more.

DeHaven, who has done this kind of work for nearly 30 years, explained why. When you monitor agents, you pick up a gut feeling for when one is “going off the rails.” He said that’s not something you can train in a day, but “something you develop over years.” He chose the experienced developer because that intuition would let him move faster than someone who had not built it yet. 

What can other multi-location brands take from Removery?

DeHaven’s advice for teams just getting started is practical.

1. Use AI in your personal life first. Recipes and travel planning teach you what you can trust an AI system with and what you should keep for yourself. He used Claude to write his trip itinerary to his calendar.
2. Be candid with colleagues. He complained about Claude’s errors when a developer asked whether he had trained it on new skills. After retraining, his project became more efficient.
3. Build your own data layer. He said the biggest difference came from building a fresh stack of APIs he controls, some read-only and some read-write, with data in JSON that an agent can read. You may build more than you need, but when new demands come up, the access is there.

How does Birdeye help lean multi-location teams?

Removery’s studios have little time for phones and follow-ups, and DeHaven’s roadmap includes AI agents answering basic calls for every studio, with the same central training and some local tailoring. The goal, he said, is that the humans have the time they need.

Birdeye supports that kind of lean-team model.

• Myna: Birdeye’s AI coworker for front office work such as calls and routine inquiries.
• Birdeye Reviews AI: The review feature DeHaven credits, where AI responds to positive feedback and routes negative feedback to a person.

FAQs