Client zero: What happened when we ran our AI-native playbook on ourselves
How do you offer clients a new way of working you haven’t lived yourself? Simple. You don’t.
At Emergn, we help companies change how they build software and products. Our global delivery organization builds solutions for clients every day, which makes the way our teams work the most visible proof of what we offer.
So, when it came to AI-native software delivery, we didn’t guess. We became our own client zero and ran the full program on our own teams with Claude, measuring the results as we went. This approach not only validated a new delivery model, but also empowered our people to experiment, develop new skills, and help define what AI-native working looks like in practice.
Where we started
At the start of 2025, AI use across our delivery teams was uneven. About half of the developers in scope were using AI tools, and demand for Claude kept growing. But every team was finding its own way. There was no shared policy, no measurement, and no way to carry what worked in one team over to the next.
Leadership wanted evidence, not anecdotes. That meant measured productivity gains with quality kept firmly under control, plus a repeatable path to an AI-native delivery model we could offer our clients. And importantly, we wanted our teams to be part of figuring out what worked, rather than being handed a new set of tools and told to use them.
Treating our teams like a client engagement
We didn’t create a separate internal program. Instead, we ran the exact program we offer customers, in three phases.
January to March was all about laying the foundations. We launched Claude Opus and Claude Sonnet on our internal AI platform, approved an AI policy, and established clear lines of accountability from the senior leadership team through engineering and practice leads down to tech leads. It may not be the most exciting work, but skip this step, and everything that follows essentially becomes guesswork.
Then came the pilot factory, which ran from April to August. During this time, four delivery teams ran controlled experiments. Developers used Claude Code for multi-file changes, test generation, and pull requests, while business analysts, QA engineers, and delivery managers used Claude and custom skills for their documents and presentations, including a branded presentation builder. Teams also built and tested agentic workflows on the Claude Developer Platform. Results from 24 source reports were consolidated into one evidence package.
In August, we turned those findings into something other teams could pick up: a scaling kit with a readiness checklist, a content hub, an assets catalog, a report on new roles, and a training and communication plan.
Progress we can back up
Every hour saved is measured net of rework. In one product team, Claude saved 500 hours across development and QA. That’s 29% of the effort, and development cycle time fell by 51%. In a controlled experiment, the same task was done four times faster with Claude. A second team saw a median saving of 37.5% per task, which freed 118 hours across six people, about three weeks of full-time work. Another team cut its one-off engineering tasks from about 136 hours to 69.
The biggest single result came from legacy modernization. Using Claude Code, a team produced a modernization plan for a platform with more than a million lines of code in about a month, against an estimate of six months. That is roughly six times faster.
People felt the difference too. Between June and July, the share of staff who said they were delivering faster rose from 62% to 76%. But speed was never our only focus. We track each team’s defect and rework rates alongside its productivity gains, and we strengthened end-to-end testing where it was needed. Faster only counts if it’s still right. And behind those numbers are teams learning to work differently. The goal isn’t simply to complete more tasks in less time. It’s to give our people better ways to approach complex problems, reduce repetitive work, and spend more of their time on the decisions and challenges where their expertise makes the biggest difference.
AI literacy for every role, not just developers
Tools are important, but their real value comes from people who know how to use them well, no matter their job title. That’s why our journey has always been about more than introducing new technology. It’s about helping our people develop the confidence, skills, and judgment to work effectively with AI, whether they’re writing code, managing delivery, analyzing requirements, or solving complex client problems.
And learning isn’t limited to formal training. Our teams have the opportunity to experiment with real challenges, share what works, learn from each other, and contribute to the practices that will shape how we work in the future.
The progress is already visible. Emergn now has over 15 Anthropic-certified practitioners and a shared library of 97 AI techniques and use cases covering more than eight roles. For delivery managers alone, we mapped 37 activities and built 10 use cases from them.
What’s particularly exciting is that this knowledge doesn’t stay with individual teams. It becomes part of how we grow together, making new skills, ideas, and approaches available across the organization.
A partnership that keeps growing
Our work with Claude is also part of a growing partnership with Anthropic that supports how we’re developing our AI capabilities and bringing them into real-world delivery. Additionally, our AI-native delivery runs on Claude Code and Claude, through the Claude Team plan and the Claude Developer Platform, using Claude Opus and Claude Sonnet alongside other models.
The number of licensed Claude users has grown from 105 to 150, and we’re heading for 270 by December 2026. As adoption expands beyond the initial four pilot teams, more people across Emergn will have the opportunity to develop their AI skills, experiment with new approaches, and contribute to an evolving way of working, supported by the evidence and playbook we’ve already built.
We didn’t want to sell clients an AI-native way of working we hadn’t lived ourselves. Running the program on our own teams with Claude gave us measured results, including development cycle times cut in half and a million-line modernization planned in a month, and an honest picture of what it takes to make that repeatable.
Andrew McDonagh, Head of Engineering Excellence & AI Transition
Those results and the lessons we picked up along the way now go directly into our client work. But they also shape how we’re growing as an organization, creating more opportunities for our people to learn, experiment, and be part of what’s next.
The tools will continue to evolve, and so will we. What matters is building a culture where curiosity, expertise, and the willingness to challenge how things are done can turn new technology into real progress.
We proved ours by going first. And we’re only getting started.
Related posts
Why leading organizations treat engagement design as a strategic capability
Consistency at scale: delivering reliable outcomes across every engagement
The real reason delivery efforts fail: Execution discipline
