The prevailing mood in software engineering has shifted from breathless euphoria to a quiet, creeping skepticism. Browse industry forums or listen to the chatter at tech conferences, and you will encounter a growing narrative: artificial intelligence coding assistants were supposed to make development exponentially faster, yet project timelines seem to be stalling. Sprints are dragging, and the promised hyper-velocity hasn't materialized.
When we sat down with Ximedes CEO, Joris Portegies Zwart, to discuss how AI is actually reshaping workflows, a radically different picture emerged. The problem isn't that AI is failing to deliver on its productivity promise, but rather, the industry is fundamentally mismeasuring what productivity actually means in the AI era.
The idea that AI is slowing teams down is largely an industry "vibe" rather than a fact-based reality. Portegies Zwart strongly refutes the notion that development velocity is inherently dropping due to an AI adjustment period or review bottlenecks. Instead, the integration of these tools has seamlessly elevated the baseline of what a development team can achieve within a standard timeline.
"We are not seeing productivity drop at all," says Portegies Zwart. "Instead of losing time to an adjustment period, our teams are using AI to unlock value and deliver unexpected, high-quality features that go far beyond the original project brief."
When we confuse the sheer speed of typing with the actual value of shipping, we distort the purpose of software engineering. Delivering a project faster is not the only metric for success. The true measure of AI-enhanced productivity lies in expanding the scope of a project and providing clients with sophisticated solutions they didn't even know they needed.
To understand how AI unlocks this new tier of value, it helps to look at how human expertise is structured. Most senior developers operate on a T-shaped knowledge model. They possess deep, specialized expertise in a few specific domains, coupled with a broad but superficial understanding of various other technologies. Historically, venturing outside that deep vertical expertise required significant time for research and trial-and-error.
"Think of an engineer's skill set as a T-shape, where they have deep expertise in a few areas and superficial knowledge in others," Portegies Zwart explains. "AI acts as a supplement to the 'armpits' of that T-shape: the areas where an engineer understands the overarching concept, but lacks the specific implementation skills to build it quickly."
This dynamic was perfectly illustrated during a recent greenfield Ximedes project. The engineering team faced a highly complex challenge involving spatial positioning along looping routes and helices. In the past, an engineer might have conceptualized a brilliant testing framework to visualize this data, but lacked the time to learn the specific syntax required to build it. Using an AI coding assistant, the team bridged that gap instantly.
By leveraging AI for the unfamiliar technical implementation, the engineers built a highly sophisticated, SVG-based test tool visualizer. Without the assistance of AI, the team would have still conceived the idea, but they simply would not have had the bandwidth to execute it alongside the core project requirements. This shift transforms the engineering mindset from wondering if something is possible to actually finishing the sentence and deploying the feature.
This evolution requires us to rethink our project success metrics entirely. AI-driven productivity should be evaluated through a new set of criteria:
While AI handles the heavy lifting of technical execution in these unfamiliar domains, human context remains the irreplaceable core of software development. AI possesses no understanding of a client's broader ambition, business intent, or the nuanced relationship management required to guide a project to completion.
A client is invariably more satisfied receiving a product enriched with unexpected, high-quality tools than receiving a standard, bare-bones delivery a few days earlier. The engineers remain firmly in the driver's seat, using AI not just to code faster, but to build better.