Every few years, a tool arrives that changes the pace of software development. Most of them arrive quietly. This one did not.
AI coding assistants, led by tools like Anthropic's Claude Code, went from pilot experiments to daily workflow inside Ximedes in a matter of months. Not because of pressure, not because of a mandate, but because the tools were solving real problems. Ximedes CEO Joris Portegies Zwart put it plainly: "It exploded." And now, the work is figuring out what to do with that explosion.
That is the honest starting point. Not a claim that AI changes everything, but an acknowledgment that it already has, and that the real question is not whether to use it, but how.
Where Ximedes Stands
Ximedes has adopted AI tools systematically. Essentially, all engineers at the company now have access to Claude Code. Active use is running across multiple projects, including GIVA, Arriva, and D-Ticket. Where customers have not yet approved AI tooling, Ximedes proactively opens that conversation rather than waiting.
The company is also investing in proof-of-concepts, building AI features into existing products to demonstrate to customers and prospects what is achievable. A recent demo to a client showed how an MCP interface on a payment portal could help customers integrate AI into their own workflows.
The Risk Nobody Talks About Enough
The obvious fear about AI in software is bad code, and that concern is real. But at Ximedes, the bigger concern is something more subtle.
AI types fast, generates volume, and if an engineer is not careful, the work changes shape entirely. Instead of writing and designing, the engineer becomes a reviewer, reading through machine-generated output to determine what is correct, what is safe, and what is actually fit for purpose. That is not a creative job. It is an administrative one.
The solution is not to use AI less. The solution is to stay in the driver's seat. Engineers at Ximedes are trained to keep the scope of AI instructions narrow and specific. Rather than asking Claude to build an entire system, they ask it to write a specific function or refactor a specific module. The engineer owns the design. The engineer owns the architecture. AI handles the syntax within those boundaries.
Claude, like all current AI tools, lacks persistent memory and long-term context. The engineer has to supply that context explicitly, which in practice forces clearer thinking about what is actually being built and why.
Quality Does Not Change Because the Tool Does
Joris Portegies Zwart is direct on this point. The speed at which AI generates code makes guardrails more important, not less. Code review, testing discipline, and security awareness remain non-negotiable, regardless of how the code was produced. The standard Ximedes holds its engineers to not shift based on how a line of code was written.
AI tools are confidently wrong often enough that treating their output with the same scrutiny as any other contribution is not optional. It is the job.
This is what separates adoption from dependency. Ximedes is using AI because it reduces friction, accelerates delivery, and opens up possibilities for what teams can demo, build, and offer. The company is not using it as a substitute for engineering judgment, because no tool replaces the intent and purpose a human engineer brings to a product.
AI does not care whether the product succeeds, fails, or ever ships at all. The engineers at Ximedes do. They carry the context, the intent, and the accountability that no model can replicate. That is not a limitation of the technology. That is the point.
