There is a lot of speculation around the topic of replacing developers, and quite frankly replacing a lot of workforce within a lot of companies with Agents. While researching for this blog post I found many articles with titles like:

• Death of software engineering
• AI will write most of the code by 2026/27/28
• AI will lead to reduction in software engineering jobs by x %

A discussion with a colleague revealed a similar thought process, although for a different job, when she said “Well for a basic analyst just doing research, I believe this can be fully done by AI.” But neither this colleague nor most of the authors of those articles have worked in the field, they have been predicting. If you do not know what the work really consists of, a statement about the degree of possible replacement with agents is, to say the least, risky. But I think everyone agrees that AI and agents are currently changing the ecosystem for software developers such as myself.

Ecosystem change

The ecosystem is made of much more than just code generation with your favorite LLM or agent. To give a few examples it changes the way

• you work in teams
  • Agile workflows are transforming e.g. BMAD, which is a framework in which Agents are collaborating with you and act as a whole scrum team.
• you handle infrastructure
  • Infrastructure handling e.g. Replit a company which recently was in the news for one of their Agents going rogue deleting the production database
• your job changes
  • with new AI tools booming on the market, it seems that each day we should restructure our working routine.
• of writing code
  • The spotify CEO for example states that some senior developer at spotify did not write a single line of code in 2026. Google states that over 25% of new google code is written by AI. But let’s take both examples with a grain of salt since these are most likely marketing messages and not technical reports we could verify.

Ok but is this change positive for us as developers? Will we be lead into a golden age of AI supported software engineering. Where we will build cool quality products in no-time or will it be prompt hell. Where nobody wants to check the generated code and bugs and outages are the norm? Let’s have a closer look at the two sides of this coin.

Pros and Cons

Pros

• Fast setup Yes with Agents you can get started in complex projects in no time. Same is true for MVP or PoCs it is easy and faster than before.
• Tool Dependency There are certain tools, which usage was hard if you were a beginner e.g. regex. Today you can simply ask an Agent to give you a fitting regex and boom there you go. No fiddling around over hours just to get working. Obviously the output might not always be best but it brings you at least 80% there. Talks like this, have aged nicely in this regard
• Less boilerplate A lot of code is boilerplate and especially for well used frameworks you do not have to define every single api route by hand but instead can supply a rough list to an agent and can adjust afterwards.
• Increased learning You don’t understand something, why it was done in a specific way or how it works? Agents will save you from calling your senior dev and then still not getting it.

Cons

• Learning by not doing One drawback is that if you stop writing code and instead write specs(as in spec-driven development) you learn to develop differently and you might not learn some things at all. Is this really bad? Only time will tell.

• Every line is a liability Agents and LLMs are particularly good at pushing out generated text but in software each line of code(yes documentation also counts) is a liability. Since AI tends to produce more code and documentation it therefore also produces more liabilities.

• Invested Time shows commitment In Pre-Agent times if someone opened a PR, created a package or build an app and spend real time on that it showed commitment. Which in turn lead other people to commit(git pun intended) as well. As example: “You have invested days in this feature, so I will invest hours to review.”. But now what if your colleague only invested 1 hour or 5 minutes, and touched 50+ files, would you review this with the same precision as before?

• Management expectations Let’s put aside the question by how much AI and Agents can boost your productivity, double it and you might be close to what your manager is expecting.

• Less fun coding In short some developers love writing code, thinking about how it all works together, how to make it efficient and learn along in the process. Just prompting an AI to do this is a lot less fun to be honest. One article going more in depth can be found here

Neutral

There are some blogs or articles stating there are security and ethical concerns regarding the use of AI. That Agents might be biased and we would not know in which direction etc. As far as I can tell yes security and biases might be a problem, but this was also the case before AI. There are enough examples of human failures in software development e.g. Accenture documents have been leaked on public s3 buckets, with no AI involved(just one example of many).

What to do now as developer?

Well here are some of the tips I would argue will help you at least lose your job slower.

1. Do what AI can’t:

• communicate with your team client or customer to get the best idea of what is needed
• Map the problem to the domain language

2. Use AI to your advantage:

• get familiar with some tools(focus on the big ones for now). Don’t let the FOMO hype get to you, not every tool needs your attention.
• use AI as teacher, if you don’t understand something that works do not just skip over it, try to understand.

3. Consider all dimensions of development

• in a great ted talk, which sadly I can not find anymore, the author said code can be optimized into different dimensions e.g. speed, maintainability, readability, documentation completeness, testability etc. With the rise of Agents and vibe coding I see a tendency to get something working but all other dimensions are lost in consideration.

It will be very interesting how development will shift in the near future. As you might have noticed I am leaning towards a conservative side of things, so take this article as always with a grain of salt. Further reads:

• AI can write your code it can’t do your job