I started as a software developer in 2008. Three snapshots taken nine years apart show how much the work has changed and where the value of an experienced developer is moving.
2008: handcraft in the server room
Applications were compiled on the developer’s own machine and installed by hand on Linux servers. Those servers could be physically located on site, for example at industrial sawmills. If a sawmill’s server suddenly stopped responding, the developer at best had to drive there in person and figure out from the machine’s command line what had broken. DevOps was an unknown term, as the first DevOpsDays was held only in 2009. GitHub had just been launched, and version control was not a given. Sometimes changes were made directly in production. Help was found on the internet, but also in books and above all from colleagues. Most of the time went into implementation and manual testing, because people trusted human work more than automated tests.
2017: everything in the cloud
Code lived on GitHub, CI pipelines built the applications and deployments went to the cloud. Many good practices still in use today became established. Help was found faster on discussion forums, in guides and on blogs than in books or from colleagues. Still, most of the time went into implementation, testing and reviewing colleagues’ code. The specifications and ideas were already in place; they just had to be built. Testing had been shifted towards automation, but testing done by people was still valued.
2026: in production by the afternoon
A customer orders a change to an application that was last developed in 2017. The codebase is old, but AI agents implement, test and deploy the change by the afternoon. At the same time, anyone can buy an agent, even someone who has never looked at a single line of code. The agent builds them, in less than a day, an application that used to take several people weeks or months.
The way we look for help has changed too. The familiar guides and blogs now focus on harnessing AI instead of specific technologies. We still read guides and ask colleagues for help, but we no longer ask how to do something with a particular technology. Instead, we discuss which AI model and which instructions give the best result.
Speed has, however, moved the bottlenecks elsewhere.
Before implementation, the bottleneck is ideation. After implementation, the biggest bottleneck is the combination of testing, code review and customer acceptance. Manual testing is, surprisingly, the same bottleneck as in 2008: we do not yet fully trust that AI can do in hours the work that would take a person several days. Code review takes more time than before, because more code is produced and what is being reviewed is AI output rather than human-written code. At its best, AI implements features very sensibly. At its worst, the codebase gradually drifts away from what the team is used to, and maintaining it becomes harder. Finally, the customer still has to accept the change, and acceptance does not speed up at the same pace as implementation.
A good example of this is a case from last summer. A relative of mine who runs his own machine maintenance company had kept his customer and service records on paper. He asked me to build a digital service book through which his customers can order maintenance online. An impressive prototype was ready in one evening, and a production-ready system after a few more days of work. It is, however, still waiting for my relative’s testing and acceptance.
The same was visible in the user interface renewal of the Finnish National Agency for Education’s ePerusteet service, which I carried out:
AI produced about 90 percent of the changes to more than 100,000 lines of code, but before deployment the agency’s experts tested the whole carefully.
Competition is decided by the idea
When anyone can build an application, competition shifts to who builds the best idea. AI produces what someone has already written at some point. A prompt like “create a customer relationship management system” or “create a platformer game” produces a working application, but anyone can get the same one. Without an idea of your own, the result does not stand out from the crowd.
On the other hand, your own idea no longer gets buried under the implementation work. A few years ago, back when AI-assisted development was still just smart autocomplete, some friends and I came up with an idea for a board game. The prototype was never built, due to lack of time and a bit of laziness. Recently I tried whether agents could turn it into a version playable on a computer. In a few evenings the game was so complete that it may actually work even better in digital form.
In customer projects, uniqueness rarely comes from inventing something completely new. It comes from the customer’s own business: its processes, its exceptions and the things nobody has ever written on the internet. My relative’s digital service book is not a new invention, but because it is built around the way his company works, it has better features than competitors’ systems. That is why technical consulting is shifting from code to understanding the customer’s business and translating it into applications.
What this requires from an experienced developer
Eighteen years of experience have not lost their value, but the way that experience is used has changed. I know what good architecture looks like, which solutions last for years and where an old system typically breaks. That lets me steer agents and evaluate their output. I also know how to ask the customer the right questions before a single line of code is written.
What this enables for the customer
When implementation is fast and affordable, the customer can try out more ideas and drop the bad ones early. The budget goes into finding the right solution rather than building a single version. The customer pays less and less for lines of code and more and more for a result that supports their business.
Are you interested in Gofore as an employer?
Grow with AI
This blog is part of Gofore’s Grow with AI theme, where we explore how artificial intelligence is changing expert work and share Gofore employees’ experiences of how their work is evolving and how they are learning new skills.
Explore more from the Grow with AI theme:
What does expertise mean in the age of AI?
How can software developers keep their skills up to date in the age of AI?
How is AI changing the day-to-day work of a software developer?
