Skip to content
All insights

Insight · Process

How We Use AI in Our Development Workflow

We are fans of technology, we're eager to learn about new tools and see how we can best use them in our workflows, so naturally we are excited to test AI capabilities and see how they can help us work better.

June 9, 2026 · 7 min read

How We Use AI in Our Development Workflow — illustration

As a team that has always enjoyed new technology, we tend to try new tools soon after they appear, and AI has been the most interesting of them in a long time. What started as curiosity and a few experiments is now part of how we design, build, review code and keep track of our work.

We believe using the right tools is always a step forward, and that a good team keeps upgrading its skills and procedures as those tools change. We use AI where it makes us faster and more consistent, while people stay responsible for every decision that matters. Below are the places it has changed our work the most, and what that means for the clients we build for.

Stable Tools and Experiments

We approach AI in two ways. The first is the set of stable tools and tested methods that are part of our everyday workflow, which we use on client projects only once we know how they behave, and whose output is always reviewed by a person before it goes anywhere.

The second is experimentation. We also like to play with newer tools, more complex automations and ideas that could change how we work, and we do that purely to learn, on internal projects kept well away from client work. When an experiment proves itself, it can move into our workflow, and until then it stays in the lab.

Learning New Tools and New Processes

Adopting AI has taken real effort from everyone on the team. Each new tool comes with its own way of working, and getting good results from it means learning how to give it the right context, how far to trust what it produces and where it fits into the steps we already follow.

That learning never really finishes, because AI changes faster than any tool we have worked with before. New models come out every few months, each one better at some tasks than the last, and new ways of using them keep appearing alongside, so a method that worked well at the start of a project may already have a better alternative by the end of it.

Quick Access to Specialist Knowledge

Projects often touch fields outside our own, such as a regulation we haven't worked with before, an unfamiliar industry or a technical problem in a narrow area. In the past, even exploring an idea like that usually meant finding and hiring an expert before the work could start.

AI lets us research these areas much faster and in far more depth than a search engine allowed. It is often enough to understand the problem, shape a first version of the idea and talk about it with the client in concrete terms. When a project needs deeper expertise, we still bring in a specialist, and by then everyone knows exactly what to ask them.

From Hand-Built Animation to AI

Animation has been part of our work for a long time. For Van Stapele, every scene began as a painting, which we separated into layers and animated by hand in code until scrolling through the site felt like stepping inside the bakery. For The Neverlands, we layered hand-painted illustrations so the artwork would move with the scroll and the mouse.

At the time, no tool could turn a painting into a scene you scroll through, so every movement was built and tuned by hand, and getting it to run smoothly on every screen was the hardest part of both projects. Today we can use AI to generate animations from artwork like this, which lets us explore more ideas in the same amount of time and spend our own effort on art direction and on how the motion feels.

Automating the Routine Work

A lot of time on any project goes into work around the code, such as writing tickets, updating their status, answering questions about progress and preparing reports. Through MCP integrations, we connect AI to GitHub, to Linear, the tool we use to track our work, and to Slack, so the same assistant that works in the code can also handle much of that routine.

It can draft tickets from the findings of a review, link a piece of work to the issue it belongs to and keep each issue's status in step with what has been merged. A project manager can ask in Slack what changed this week, what is blocked or what is ready to test, and get an answer drawn from the actual commits, pull requests and tickets.

For clients, this means less of the budget goes into coordination and more of it goes into building the product. Status updates take minutes instead of hours, the project board stays accurate without anyone having to update it by hand and developers spend their time on the work itself.

Building Trust in the Code

Every change we open now gets an automatic review before a developer looks at it. The AI reads the change in the context of the whole codebase and points out likely bugs, unhandled edge cases, and accessibility or layout issues that are easy to miss after hours spent on the same code.

Our reviewers still read every change, but with the smaller problems already flagged they can spend their time on the bigger questions of design and architecture. Issues get fixed while the work is still fresh in the developer's mind, rather than weeks later when they turn up in testing, and over time the codebase becomes easier to maintain and safer to change.

The same approach helps when we take over an existing product, where the code may have been written by several teams over several years and the documentation is often thin. AI helps us scan the repository, trace how data moves through the system and surface areas that need attention, such as outdated dependencies or security concerns, so clients get a clearer and faster assessment of what they have.

Keeping Products Up to Date

AI has also become part of how we maintain products after launch. Every product depends on third-party packages, such as the NPM libraries a web app is built on, and keeping them current is one of the most important and most time-consuming parts of maintenance.

We use AI to check which packages are out of date or have known vulnerabilities, read through their release notes and update the code that a new version breaks. A developer still reviews and tests each update before it is released, but the slow work of tracking changes across many packages is done much faster, so products stay secure and clients spend less on keeping them that way.

We Keep Full Control

However keen we are on new tools, the AI only makes suggestions and people make the decisions. Every change is still reviewed and approved by a developer, and nothing ships without a person signing off on it.

We take the same care with privacy, and for our healthcare clients, we pay close attention to which tools can see which code and data, and we keep patient information out of places it shouldn't be.

Final Thoughts

The tools will keep changing, and we plan to keep learning with them. Each time we find one that helps, we adjust our process around it and share it with the rest of the team. For our clients, that means fewer issues reaching production, faster answers when they ask how things are going and more of the team's time spent on work that needs real judgment.

If you're curious how this would work on your project, get in touch and we'll walk you through it.