Minor changes

Author: marcusoftnet

_posts/2025-08-04-ai-scare.md

@@ -15,33 +15,35 @@ Because reading about AI, and it's application in digital product development, i
 
 I do think that generative AI is a paradigm shift, but just as with others that I've experienced (Internet, Open Source, Agile, Cloud, Serverless and DevOps to mention a few) it is hard to know where it will take us, how it will change us before the dust has settled.
 
-I stay positive though and think that the future for digital product development is great, engaging and fun. Also, as with many paradigm shifts human jobs have evolved rather than been replaced. We are moving up in the value chain, rather than becoming obsolete.
+I stay positive though and think that the future for digital product development is great, important and fun. Also, as with many paradigm shifts human jobs have evolved rather than been replaced. We are moving up in the value chain, rather than becoming obsolete.
 
 Let me clarify my thinking a bit.
 
 <!-- excerpt-end -->
 
 (There will be some "might", "potentially" and "could" in this post. I don't claim to know anything about how it *actually* will be. I'm just trying to play out some scenarios and see where that future might take us.)
 
-Digital product development has always been a complex adaptive endeavor. We don’t start with certainty — we start with hypotheses. What customers will use, value, or even pay for is something we only discover by experimenting, putting something out there, learning from feedback, and iterating. This is why iterative and discovery-driven development has always been the heart of good product work. This has not changed, and as long as there are humans in the loop (users for example) - I doubt it will change going forward either.
+Digital product development has always been a complex, adaptive endeavor. We don’t start with certainty — we start with a hypothesis or idea. What customers will use, value, or even pay for is something we only discover by experimenting, putting something out there, learning from feedback, and iterating. This is why iterative and discovery-driven development has always been the heart of good product work. This has not changed, and as long as there are humans in the loop (users for example) - I doubt it will change going forward either.
 
-Now, in the age of generative AI, one part of this equation has changed dramatically. AI has given us a tool that can, potentially, accelerate certain tasks—particularly coding. It’s not going to replace software developers, but it may replace “typists”: those whose role is limited to translating requirements into syntax.
+Now, in the age of generative AI, one part of this equation has changed dramatically. AI has given us a tool that can, potentially, accelerate certain tasks, particularly coding. It’s not going to replace software developers, but it may replace "coders": those whose role is limited to translating requirements into syntax.
 
 In this [great video Dave Farley](https://youtu.be/fcjBfSiyI0k?si=rrnzzIHelSGs0qEA) makes the distinction between "coder" and "developers". I think that this distinction sits at the heart of understanding how the developer profession might change in the future.
 
 But this shift creates a deeper, and more interesting, question: if code could be generated faster than ever, where does the real bottleneck in product development move?
 
 ## Code Has Always Been a Cost
 
-As Dan North puts it: “Code is cost.” It costs money to write, but even more to maintain. AI might reduce the cost of producing the first version of a feature, but it doesn’t remove the need for good practices, architecture, testing, security considerations, and integration with existing systems.
+As Dan North puts it: [“Code is cost.”](https://dannorth.net/blog/cupid-the-back-story/#open-closed-principle) It costs money to write, but even more to maintain. AI might reduce the cost of producing the first version of a feature, but it doesn’t remove the need for good practices, architecture, testing, security considerations, and integration with existing systems.
 
 Some futurists imagine a world where code could be “regenerated” from scratch whenever requirements change—skipping maintenance altogether. Maybe one day. But today, software isn’t just code; it’s entangled with data, APIs, compliance constraints, and business context. AI can help us write software faster, but it doesn’t erase complexity.
 
+We do not version control binaries (anymore) - we version control high-level source code that compiles to the binary. I don't see a future where we will not keep track of the AI generated source code and just keep track of the prompt. Natural language is just too vague to be able to express what we want the system to do.
+
 ## The Real Bottleneck Has Shifted
 
 If AI is making coding faster, then what’s slowing us down now?
 
-It’s not typing (Dave Farley again: `the keyboard is not the bottleneck`), I can tell you. It never has been and probably never will. More to the point, in the age of AI, writing the code the first time is just a small part of the overall work that is done to the code. I remember sayings like `Code is read more than it is written` that is true for any application that is maintained. One might wonder how that will change when tools can explain and maybe even rewrite code for us - but I still think that optimizing for writing code faster is a trap.
+It’s not typing (Dave Farley again: `the keyboard is not the bottleneck`), I can tell you. It never has been and probably never will. More to the point, in the age of AI, writing the code the first time is just a small part of the overall work that is done to the code. I remember sayings like `Code is read, more than it is written` that is true for any application that is maintained. One might wonder how that will change when tools can explain and maybe even rewrite code for us - but I still think that optimizing for writing code faster is a trap.
 
 Also, the jury is still out of on the question about how far AI tools will take us in generating great code. I really cannot say anything about how this will play out. But I'm convinced that optimizing for generating a lot of code faster will not make our product better. That comes from something else. .
 
@@ -67,6 +69,12 @@ Generative AI doesn’t diminish that mission. It amplifies it:
 
 Developers who combine technical skill with product thinking will become even more critical in this AI-driven world.
 
+## Expressing what we want
+
+As I wrote, and as [described better here](https://youtu.be/MUol_5xvFTI?si=9KeNRTS0T6Tkao2D) natural language is just too vague to describe how we want an application to work. Even on an outcome oriented level.
+
+A bit more formality is needed to express what we want, something like an Domain Specific Language (DSL) for testing like [Gherkin](https://cucumber.io/docs/gherkin/), might be a way to ensure that we get closer to what we want. But also that we get some way to verify what was built actually does what was expected.
+
 ## Why Agile Matters More Than Ever
 
 From this it follows naturally that AI doesn’t make agile practices less relevant — it makes them more essential.