Really enjoyed the topic. I agree that it has completely changed the game. After years as a professional software developer, AI coding is radically different, I can do more, have more tests and cover more edge cases. I think of it as a sort of higher level abstraction, instead of thinking in terms of classes or functions more like full functionality.
On a personal note, there is some discomfort when a model spits out hundreds of lines of code, and I go from slowly building a system which I fully understand to one with areas of increased sophistication / complexity, that is almost certainly better than I could write myself but struggle to understand.
Oh trust me, I absolutely understand the discomfort. Nowhere in this note did I indicate that this is fun or easy. Amazing and fascinating, but that's a long way from "fun".
I've had a very similar experience. I really enjoy the accelerated flow from letting AI produce stuff I already know and understand. The gap between initial idea and having something working is much faster, which is satisfying.
There's certainly a tipping point though. As you said, there becomes a point where things have gone down a weird route and it's difficult to recover. I find this to either be that it's used a slightly weird architecture, or it's way over-complicated the implementation by using uncommon language/framework features.
Also, to me it sometimes feels like I'm constantly doing peer review. Which I sort of am I suppose. I feel like making that decision of what to write yourself and what to generate is part of the process of getting used to this new way of working.
Agreed, I have thrown away hours of work, because something started well then went south and became more work to fix. You never quite know what you are going to get, a very concise understandable change, or a complete refactor with so much flexibility that it is harmful.
I liked the article and subscribed for more. Thanks for writing it up!
I've been a sw eng for 25 years professionally (35 years incl. coding at home), and have used GH Copilot from the beginning (also tripled my velocity, so I agree with you there!), Claude Code too, and built a production-grade coding agent tool for a year. I immensely enjoy this new tech.
However, I feel that right now, my $150/h wage is still justified because the tooling in its current state still very much needs my contribution to oversee what it does, as in, it's still a tool, not something that I feel have replaced my work. I do think this might absolutely change as soon as in the next year (or not, hard to tell), just wanted to say that for now, I think the change I heard you describe hasn't happened yet.
Dave I am impressed. I am also 85 and not easily impressed! I will email you with more details, but make this visible for others. You hint at requirements and architecture, as a front end for AI software generation. I agree. This is my speciality. I have invented a language and processes, which can be used to generate systems (not just code), Planguage, and I am using them with AI. See my writings on https://www.researchgate.net/profile/Tom-Gilb or in the References of Value Improvement
https://tinyurl.com/GilBot [Grok] My contact is Tom@Gilb.com. Author 1988 of Principles of Software Engineering Management and many other writing before and after
This resonates completely with my experience these last 6 months. I'm looking forward to the articles to follow, and how that aligns with my own journey.
Bravo, looking forward to following along for more. Really enjoy your writing style!
Really enjoyed the topic. I agree that it has completely changed the game. After years as a professional software developer, AI coding is radically different, I can do more, have more tests and cover more edge cases. I think of it as a sort of higher level abstraction, instead of thinking in terms of classes or functions more like full functionality.
On a personal note, there is some discomfort when a model spits out hundreds of lines of code, and I go from slowly building a system which I fully understand to one with areas of increased sophistication / complexity, that is almost certainly better than I could write myself but struggle to understand.
Oh trust me, I absolutely understand the discomfort. Nowhere in this note did I indicate that this is fun or easy. Amazing and fascinating, but that's a long way from "fun".
I've had a very similar experience. I really enjoy the accelerated flow from letting AI produce stuff I already know and understand. The gap between initial idea and having something working is much faster, which is satisfying.
There's certainly a tipping point though. As you said, there becomes a point where things have gone down a weird route and it's difficult to recover. I find this to either be that it's used a slightly weird architecture, or it's way over-complicated the implementation by using uncommon language/framework features.
Also, to me it sometimes feels like I'm constantly doing peer review. Which I sort of am I suppose. I feel like making that decision of what to write yourself and what to generate is part of the process of getting used to this new way of working.
Agreed, I have thrown away hours of work, because something started well then went south and became more work to fix. You never quite know what you are going to get, a very concise understandable change, or a complete refactor with so much flexibility that it is harmful.
I liked the article and subscribed for more. Thanks for writing it up!
I've been a sw eng for 25 years professionally (35 years incl. coding at home), and have used GH Copilot from the beginning (also tripled my velocity, so I agree with you there!), Claude Code too, and built a production-grade coding agent tool for a year. I immensely enjoy this new tech.
However, I feel that right now, my $150/h wage is still justified because the tooling in its current state still very much needs my contribution to oversee what it does, as in, it's still a tool, not something that I feel have replaced my work. I do think this might absolutely change as soon as in the next year (or not, hard to tell), just wanted to say that for now, I think the change I heard you describe hasn't happened yet.
Dave I am impressed. I am also 85 and not easily impressed! I will email you with more details, but make this visible for others. You hint at requirements and architecture, as a front end for AI software generation. I agree. This is my speciality. I have invented a language and processes, which can be used to generate systems (not just code), Planguage, and I am using them with AI. See my writings on https://www.researchgate.net/profile/Tom-Gilb or in the References of Value Improvement
https://tinyurl.com/VIpdfFREE
July 4 2025, 20 page core book, + Appendix, and References
You will notice I am using AI to generate systems planning. I suggest we must enlarge our scope to 'systems' not merely code (a sub-optimization).
See recent Gilbot, https://tinyurl.com/GilBotGPT
https://tinyurl.com/GilBot [Grok] My contact is Tom@Gilb.com. Author 1988 of Principles of Software Engineering Management and many other writing before and after
This resonates completely with my experience these last 6 months. I'm looking forward to the articles to follow, and how that aligns with my own journey.
Fantastic read. So much of this hit home. Thank you.
Being a beginner in this field, I feel like all my hopes have disappeared.
Yeah, I absolutely understand. I offer some partial answers to that here: https://davegriffith.substack.com/p/the-apprentice-problem .