AI isn't perfect but we would never want to go back to doing it all ourselves

blog@dws.team
November 11, 2025
about 1 month ago
AI isn't perfect but we would never want to go back to doing it all ourselves

For some, using a coding agent is frustrating, tiring and not nearly good enough. For us, it’s a path to becoming an all-rounder.

We're a small software development company based in Amsterdam. We do custom applications in Django. Hence our name, Django Web Studio. We also do frontends, such as the frontend app for our inspection tool.

And, increasingly, we use coding agents.

From the moment I started programming, the very first projects I did were conceived, designed, developed, deployed, maintained and managed by just one person. Me.

Think a server humming away next to my desk.

It was only after we started our company that I realised programming was actually split up into myriad roles: backend developers write in Python, Java, or C++, frontend devs know everything about HTML, Javascript and CSS, and about React, Nextjs, Vue. Then there are those who tackle the complexities of mobile apps. And DevOps.

What really hit home was that nobody we hired could be made responsible for the complete cycle. They did their bit and went home.

We need all-rounders. We'd love to have designers who are also software developers, DevOps people who crave frontend work, project managers who come up with brilliant ideas. We've been searching for years. And now we have found them, in coding agents.

We think.

Our company has a good reputation. We tend to keep clients for years. But sometimes they leave.

And then they come back.

So too this example. Client had a new frontend built by a crew who had total disregard for twenty years of clients’ history. Client saw what had become of the application and cried.

Back with us, now we've the staunch task of bringing back the old while retaining the new. We use coding agents to help us refactor our backend for the new frontend, and to refactor the frontend to use the refactored backend.

Turns out though, much of the frontend wasn’t finished. What happens when you ask a coding agent to finish the work?

Nothing good. The agent delivers code that works but UI is way off the mark. Whole swaths of code disappear. Calls are made from each component, resulting in hundreds of requests to backend.

We spend hours fixing stuff. But we learn. We learn to keep it small, to specify exactly which file, which issue, which outcome we expect.

And a strange thing happens while we learn. You see, for a coding agent, there’s no difference between backend or frontend, between your local environment and a server, between working on an app, a desktop application, or a web application. And gradually, while working with the agent, the borders between the specialisms blur. For the human. They become all-rounders.

For some, the coding agent will never be capable enough. They might experiment for some time, but then go back to their old ways. But for our dedicated team of experts, they learn to cope with the drawbacks and fickle failures of programming with AI. And from being experts, become all-rounders. And we really need all-rounders.