5 reasons we developers misjudge agentic software engineering

Throughout 2025 a kind of trench warfare raged between software developers on the pro and anti-AI development camps. We are, by definition, the experts on software creation. Ironically, this also makes us highly biased, and is exactly the reason you can't trust us to evaluate whether you should bet on things like vibe coding, AI assistance and agentic engineering flows.

This article is a call to consider the factors behind what you read and to help you question your own biases.

Natural resistance to being pushed around

In many companies, leadership pushed developers into using genAI. Irrespective of whether using genAI is a good idea, being pushed into something creates a certain amount of natural resistance in people. I don't want to be told what to do, least of all by those management guys that have no idea about technical matters!

Disappointment from unmet expectations

The hype around genAI, combined with certain impressive demos or personal experiences, can easily lead to high expectations. High expectations are hard to meet and cause disproportional disappointment when things don't work well. Most of us have watched skilled pianists and learned a simple tune on a piano fairly easily, but starting to truly learn piano brings a hard reality check. The fact that you thought it would be easy makes it much worse.

Wrong measure of correctness

Like any humans, we developers see things from our current perspective. Our current process of software development defines "correct". What an AI does is measured against this. You see an example of this whenever someone describes AI-generated code as "low quality" or refers to AI as a "junior developer".

What a second. What is "high quality", actually? What does a "junior developer" do, or not do? It's not that these measures are right or wrong, it's that these are completely artificial concepts. They are part of a framework of thinking which was invented and very successfully used by software engineers over many years, and now we find it hard to see the world from any other perspective.

The truth is, the market defines correct. If AI generates software which fails every single traditional measure of quality or completely bypasses our normal development process, and yet creates more value for customers, then customers will use that product.

Users don't care about your internal processes, they just want to solve their problem. Before you protest: "value for customer" includes quality attributes like "being available when I want it" or "not compromising my security" to the extent that they are desired by the customer.

Existential threat

I've spent 20 years building up my skills in creating great software. This is tied to my self-worth, my position and my salary. Any technology which threatens to let novices with little experience do this is most certainly bad. What's more, the idea that programmers like me may one day not be needed at all triggers me to defend myself against an existential threat.

Divas don't get their hands dirty

On the whole, software developers have it good. We get paid a relatively high amount to do work which is so enjoyable, we keep working for free in the evening in the form of open source, just for fun. Some of us get some very nice benefits like free coffee and food and fitness clubs and company cars and fancy technical equipment and work-from-wherever. If we're not treated the way we expect then we go and get a job somewhere else, where they give us the respect we deserve, baby, and it's been like this for 50 years.

Even if the above paragraph exaggerates the situation for most developers, the fact is that we are wholly unprepared for the idea of working in an unpleasant way because we have to. Other people out there work long hours shifting bricks around on a building site, not because it suits their preferences, but because that's what pays the bills. When someone tells a developer that he should use AI for economical reasons, he may simply refuse. He prefers writing code by hand. Quite frankly, we are spoilt brats. Now I'm not complaining, in fact I want to keep thing just the way they are, but am I being objective when I say that this you shouldn't use this AI thing which cramps my style?

So who can I trust?

If experts are biased, then who can you trust? This is a problem humans have struggled with for a long time in other fields. There are no really good alternatives to the knowledge of hands-on doers of the work. As a result, today you regularly take advice from people who have a stake in the outcome - medical practitioners are one example.

You are going to have to listen to (other) software engineers. At the same time, try to consider their biases. The factors listed above are not necessarily good or bad, they are just things that we should be aware of when evaluating opinions - including when questioning our own.

What do you advise?

"Would you trust a coal miner to advise you on whether to transition away from coal-fired power?" I asked my colleague Felix Medam last week. "As software consultants, are we the coal miners?"

I liked his answer. Yes, we are the coal miners, but only if we behave that way and keep our blinders on. If we manage to set aside our biases and master both approaches, we can act in the role of the neutral energy advisor – advising and deploying solutions that provide the desired business value needed by the client, irrespective of our traditional technical preferences. And that's what I try to do, in this article and otherwise."

You too carry the baggage of biases. But you can work against the above points - choose rather than being pushed, drop the expectations and let the experience speak, optimise for business value as your measure of correctness. If you manage all that, you will have a lot less existential threat to your diva status to worry about, girl 💅.