Thoughts on AI and software development - Part 1

This blog series will be a bit different. It might leave you with more questions than answers and I apologize for that upfront. It is also more controversial than most other posts, I have written before. In this post, I will discuss the current developments regarding AI in software development more from a CTO’s perspective. This means, I cannot simply reason about the pros and cons of a topic but I also need to take the market forces into account.

This requires to also take an unembellished look at some of the darker sides of our industry, a side I usually only slightly touch in my other posts. I do not know how comfortable you will feel with this perspective but I hope it gives you some food for thought. Maybe, it even provides you with some new and useful perspectives. We will see …

Before we start: As this blog post (again – sigh!) has become way too long for a single post, I decided to split it in 4 parts:

1. A projection of a future and why we should ponder it (this post)
2. The forces that drive the markets and decision makers (link will follow)
3. Short- and mid-term consequences and some unresolved questions (link will follow)
4. How to hedge our options and move on (link will follow)

Let us begin with a bit of introduction to understand the rest of the post series better.

The role of a CTO

I wrote, I will ponder a topic from a CTO’s perspective – which immediately raises the question: What is a CTO’s perspective? To be honest: I think, this depends massively on the company, a CTO is working for. Thus, my description here cannot be generalized. It is more of a “personalized” description. On the other hand, this is fine because we are just setting the stage for the rest of this blog series.

With that being said: My business card states my job title is “CTO”. The company I work for is an IT service provider, mainly doing software development and consulting related to software development (there is a bit more we are doing but this is not relevant here). This makes the role a bit special. In IT product companies and non-IT companies, a CTO usually defines the technology strategy (let us omit the startups where the CTO more often than not is that one founder who knows how to write code). It is not that straightforward if you work for an IT service provider because in the end, our clients have the say when it comes to their technology strategy.

Nevertheless, also as an IT service provider we need to make sure that we have the right topics in our shop windows and send out the right messages. It is a competitive market and if you want to differentiate yourself you need to address the right topics at the right time ¹. This is where my role comes into play. In very simple words, my job is to advise our board regarding our general technology strategy, i.e., which topics we need to put a focus on now to be in the right spot in the future (there is a bit more to the role but then again, this is not relevant here).

Again simplified, this means I observe what is going on on the market, which topics may gain traction in the future and how they can affect us and our clients. I try to understand the pros and cons of the topics and try to anticipate how the market will most likely respond to those topics. Especially the latter can be challenging because opposed to what someone might have told you in your economics class the market rarely acts rationally.

Or to be more precise: The market does not act according to its needs but to its wants and other short-term incentives. I wrote about the need-vs-want dilemma in a prior post. Additionally, most decision makers (who ultimately define the market response) hate risk. They prefer the well-trodden paths even if they know in the back of their minds it does not solve any problem. But you do not get fired for doing what is generally accepted as “best practice”. And then there are people who play with these response patterns to benefit their individual agenda. As a result, the market often reacts very irrationally.

This makes it a lot harder to decide which recommendations to present to your board. While it is intellectually quite simple to understand which problems a topic solves and which it does not, it is a lot harder to anticipate how the market will respond to it. It is especially hard if you know the topic will not solve any problem and most likely still will create high demand. But that is the situation I am in and I try to give my board the best advice I can.

In the remainder of the post, I will lay out parts of the thoughts and reasoning, I go through in this role. Having written this, let us jump in without further ado …

A projection of vibe coding

The trigger for this post series was a recent blog post by Steve Yegge, titled “Revenge of the junior developer”.

If you do not want to read the whole blog post, here is a quick summary created by GPT-4o mini which should be good enough to give you at least a rough idea what the post is about:

The blog post titled “Revenge of the Junior Developer,” authored by Steve Yegge, discusses the evolving landscape of software development driven by AI, particularly focusing on the concept of “vibe coding.” This term describes a new era in programming where developers leverage AI-driven coding assistants to automate much of the coding process.

Key Points

1. Vibe Coding and its Evolution:

   • Vibe coding refers to a relaxed approach to coding where developers let AI handle most of the work, especially for prototyping. *The post outlines the different waves of coding methodologies, including traditional coding, AI completions, chat-based coding, and the emerging use of coding agents expected to revolutionize the industry.

2. Transition to Coding Agents:

   • The introduction and rapid adoption of coding agents are described as transformative. These agents can handle complex tasks with minimal human intervention, freeing developers to focus on higher-level work.
   • There’s a forecast that coding agents will soon surpass chat-based coding in effectiveness.

3. Financial Implications:

   • The cost associated with using coding agents can be substantial, likening their operational expenses to hiring junior developers.
   • Companies will need to budget accordingly to stay competitive, as those who invest in these technologies are expected to thrive.

4. Revenge of the Junior Developer:

   • The post compares the adaptability of junior developers to the more cautious approach of senior developers, suggesting that the former are embracing AI technologies faster.
   • Junior developers are noted to be more enthusiastic about leveraging AI in their work, which positions them favorably in the evolving job landscape where coding by hand may become less common.

5. Call to Action:

   • Yegge encourages all developers to embrace coding agents and AI tools to maintain relevance in the industry. He emphasizes the need for ongoing education in AI-driven methodologies to thrive in future software development environments.

Conclusion

The blog paints an optimistic yet cautionary picture for the future of software development, encouraging adaptation and forward-thinking among developers to harness the power of AI effectively. It highlights a potential shift in job roles within the industry, where the focus may shift from hand-coding to managing AI-driven coding workflows.

You still may want to read the whole blog post to get all the details and nuances of the post. And as Steve has a very entertaining way of phrasing things, it is even fun to read.

A grain of salt, or two, or three …

Before we jump to any conclusions and hectically hand over code creation to the next AI agent, let us make clear that Steve currently works for Sourcegraph, a company having written “Industrializing software development with AI agents” in bold letters over their home page. Thus, we better take Steve’s post with a grain of salt … or two … or three … the heck with it, pass me the whole bag!

I mean, we could also interpret the post like this:

Hey, I work for a company trying to make big bucks with AI agents for software development. First of all, I do not bite the hand that feeds me. And second of all, part of my competitive salary depends on the revenue of my company. Therefore, let me use my reputation to feed the hype and suggest company budget owners to reserve lots of money they can spend on our products in the near future (at the expense of senior developers they unfortunately need to fire to pay our bills).

Personally, I doubt that Steve’s motivation for his post was that mundane. I rather think, he is really convinced that things will happen the way he described them. But even if we talk about Steve Yegge here, we cannot assume that he was completely unbiased when he wrote his post.

A possible future

However, even with all potential conflict of interests and resulting bias in the post, it outlines a possible future of software development. I cannot rate the exact probability of this future becoming reality. But it definitely is a potential future and at the moment, a lot of people are throwing lots of money (we talk about billions, if not trillions of dollars) on increasing the probability of this potential future becoming reality.

Yes, we see signs that the AI bubble may burst soon. Some of the high-profile players reduce their investments. We also see pioneers like Yann LeCun moving away from GenAI as they do not see the future of AI there. As Yann just said in the session “Frontiers of AI and Computing: A Conversation with Yann LeCun and Bill Dally” at the NVIDIA GTC 2025:

“I’m not interested in LLMs anymore - they’re the past. The future is in four more interesting areas: machines that understand the physical world, persistent memory, reasoning, and planning.”

Still, if we look at the market, the GenAI hype is unbroken. It is still the topic of virtually every IT-related conference. Every media outlet and IT-related magazine writes about GenAI back and forth. Every new product has “(Gen)AI built-in” (even if nobody asked for it). Companies implement GenAI solutions everywhere. And interestingly (I will discuss later why I find it interesting) most developers try to surpass each other with how much GenAI support they use for their software development.

Thus, even if the forerunners prepare for the burst of the bubble, the hype is unbroken. Additionally, there are still so many bets on GenAI running that the bettors will do everything to avoid or at least postpone the burst of the GenAI bubble. Hence, the projection of Steve is a possible future. This future would have a massive impact on the business model of any IT service provider that makes a big part of its money with software development – which means I cannot simply ignore it in my CTO role.

Of course, we could dismiss the whole idea with some ironic reference to the employer of Steve Yegge, labeling him as an AI fanboy because he needs to be one. We could also dismiss the whole idea with pointing out that writing code is maybe 5%-10% of the work of a software engineer and that the real work is not writing some piece of greenfield code but to integrate new business requirements into huge existing codebases which is a completely different game. We could point at the hype industry, at all the investors who need to create massive demand in a place where natural demand normally would be average at most – and how they have perfected pushing the media’s and decision makers’ buttons. And so on.

I am also cautious regarding all those overly hyped promises of AI and have already written about it (see, e.g., my blog series about the fallacies of software and what it means for AI or my blog series about the hype industry in general and its detrimental effects).

Hence, I could leave it here. Everything has been said. From the perspective of the owner of this blog, this is true. From from my CTO role’s perspective, it is not that easy.

First they ignore you …

It is very tempting to dismiss all those “AI enthusiast” scenarios with some snarky, sarcastic remarks – and sometimes I feel tempted, too, even in my CTO role.

But then comes a quite famous – and oftentimes true – quote to my mind:

“First they ignore you. Then they laugh at you. Then they attack you. Then you win.”

Or the a bit less aggressive variant:

“First they ignore you. Then they laugh at you. Then they attack you. Then, it is considered self-evident.”

It is not clear who the originator of the quote is. But maybe this is already a hint at the universal applicability of this quote – especially if we talk about disruptive change, about change where the profiteers of the future status quo are different from the current ones.

We currently see a lot of ignorance, ridicule and aggressive backlash when it comes to AI, and as I have written before, I am also quite wary regarding all the totally overheated hype and inflated expectations regarding AI.

However, this is the moment when my inner observer reminds me of this quote and points out that maybe I could be one of the persons the quote describes – that a disruptive change might be on its way and I could one of the persons who do not want to see it until it is too late.

Stupid observer! But … well … he has a point, I am afraid.

This could be a disrupting change that is on its way. And in my CTO role, I cannot ignore it.

As I wrote before: Part of my current job is to understand possible future developments of the IT market (to the degree it is possible to anticipate such things) and support my company in positioning itself regarding those developments and their likelihoods.

Additionally, it is also important for me personally to hedge my own future career options – nothing is worse than to find oneself suddenly on the “obsolete” shelf, especially in my age. Well, I guess, we all need to be a bit of our own CTO who proactively takes care of our career options.

Pondering the possible future

So, I started to ponder the future, Steve Yegge described in his post. I cannot accurately rate its likelihood as the wannabe profiteers of this future are so busy fueling the hype with one “groundbreaking breakthrough” announcement after the other that it has become impossible to evaluate the underlying potential thoroughly. Maybe this is the plan of those people, to keep everyone breathless and prevent us from evaluating things realistically.

We have seen a similar hype with the Blockchain bubble which left a very bad aftertaste because we eventually realized that it was exactly this: A hype. A bubble. Solutioneering. A solution without a problem, artificially inflated.

However, the hype we currently see with GenAI, is at least a dozen times the size of Blockchain. There is so much blah-blah, exaggerations, hyperlatives piled up, higher than any skyscraper, and every day we are bombarded with the next hundred “mind-shattering, groundbreaking breakthroughs” that it became impossible to dig through this pile of delusion to the facts below all this.

But even if I cannot accurately determine the likelihood of this possible future, I need to ponder it …

A probably unneeded future …

So, do we need the future, Steve sketched?

The answer to this question is the simple part: No, not at all.

As I discussed in several places (see, e.g., my “Forget efficiency” post, my “Responsible IT” posts or my “Simplify!” blog series), the prevalent problem in IT is not that we would not write enough code. Quite the opposite: We create millions of lines of code every day, nobody needs nor wants – pure waste that only amplifies our actual problems.

In our efficiency obsession, we ignore everything that would be really important these days, like dealing with the increasing unpredictability of future developments at all levels ², the ever-growing complexity of our system landscapes, ridiculing today’s indispensability of a dependably working IT.

Instead, we press hard increase our efficiency, this way creating extremely rigid and (unintentionally) wasteful organizations and processes that hardly produce anything useful – simply because they are optimized for a reality that no longer exists (see the aforementioned posts for more details).

As all those efficiency optimizations do not show the effects, we expect them to have, we press harder and harder, adding more and more “of the proven” which only makes everything worse. We are pushing to do the wrong things better and better which reminds me of a quote of the brilliant Peter Drucker:

“There is nothing worse than doing the wrong thing well.” ³

Thus, if we briefly ponder our actual problems and then look at the future, Steve projected, we immediately see that it does not solve any problem we currently suffer from. Quite the contrary, it only makes things worse. It would make sure that more and more code is written in less and less time, that we not only create millions of lines of code every day, nobody needs nor wants, but billions of them. Waste to the square. Problems cubed.

And all this does not yet take the ecological footprint of such a GenAI based solution into account.

In short: The proposed future does not solve any of our actual problems in IT. Instead, it reinforces the existing problems.

Can we thus conclude, we can dismiss this potential future?

Well, unfortunately no.

At least, in my CTO role I cannot dismiss this future that easily. As I wrote in the beginning of this post, the market rarely acts based on its needs. Hence, I am not yet done with my pondering.

First interlude

We started with Steve Yegge’s post where he made a projection from the vibe coding of today to controlling AI agent fleets in the near future that take over all coding reliably. As Steve works for a company that implements exactly such an agentic AI software development product, it is easy to feel tempted to dismiss his whole projection, maybe even to ridicule it.

But as all disruptive changes start this way (anyone still remembers how much fun people made of the first digital cameras, the first iPhone or AWS’ initial cloud offerings?), I need to ponder Steve’s projection as a possible future in my CTO role.

We quickly saw that this future is not what we need as our actual problems in software development are not due to too little code output but lie somewhere else. But even if not needed, we cannot simply dismiss this projection because we know that market forces and decisions are not necessarily driven by what is needed but what is wanted.

Therefore, we need to explore the want side in the next post (link will follow). This is where it will become a bit darker and more controversial. However, rose-colored glasses do not help if we try to realistically ponder the given projection and its likelihood. Stay tuned …