Personally, I think this workflow is a trap for experienced developers, and a complete collapse of expertise for those that don't already have decades of experience to have cultivated the said taste and judgement you refer to.
Coding is not drudgery; it's a unique blend of logic, planning, troubleshooting, critical thinking, creativity, and sometimes even math. If you're experienced, you will eventually stagnate and stop growing, because its the friction from coding that cultivates ongoing learning.
For everyone else who hasn't been building for decades, it creates an interesting paradox: If these tools demand expertise, yet the tools can actively circumvent the friction that cultivates expertise, then what is the path for one to become an expert so they can effectively use these tools?
I'm not saying we need to type everything by hand, that world ended decades ago with various tooling and workflows, but the notion of not caring about the details and diving into code at all seems like the same kind of hubris that drove UML as the end of needing to care about how an application gets constructed. Turned out: complexity compounds and there's still a deep need for understanding and caring about the details of how something gets built.
Your PHP optimization loop is an interesting example, because to me, that was a failure on your role as a learner; using the tools in a Socratic manner and leaning on their "interactive documentation" aspects would have yielded the exact same effect; you could have inquired "These are my specs and requirements; what is the best platform to build this in?" The tools are amazing at data distillation and synthesis, and pattern interpolation. Maybe you would have even received other suggestions that were even better than PHP (like Go). My point is: your agent didn't "decide", it searched and found a single result. And without digging into the details, you don't actually know if that was, indeed, the best option.
Personally, I think this workflow is a trap for experienced developers, and a complete collapse of expertise for those that don't already have decades of experience to have cultivated the said taste and judgement you refer to.
Coding is not drudgery; it's a unique blend of logic, planning, troubleshooting, critical thinking, creativity, and sometimes even math. If you're experienced, you will eventually stagnate and stop growing, because its the friction from coding that cultivates ongoing learning.
For everyone else who hasn't been building for decades, it creates an interesting paradox: If these tools demand expertise, yet the tools can actively circumvent the friction that cultivates expertise, then what is the path for one to become an expert so they can effectively use these tools?
I wrote about both of these situations in-depth:
https://larsfaye.com/articles/agentic-coding-is-a-trap
https://larsfaye.com/articles/ai-coding-will-prevent-expertise
I'm not saying we need to type everything by hand, that world ended decades ago with various tooling and workflows, but the notion of not caring about the details and diving into code at all seems like the same kind of hubris that drove UML as the end of needing to care about how an application gets constructed. Turned out: complexity compounds and there's still a deep need for understanding and caring about the details of how something gets built.
Your PHP optimization loop is an interesting example, because to me, that was a failure on your role as a learner; using the tools in a Socratic manner and leaning on their "interactive documentation" aspects would have yielded the exact same effect; you could have inquired "These are my specs and requirements; what is the best platform to build this in?" The tools are amazing at data distillation and synthesis, and pattern interpolation. Maybe you would have even received other suggestions that were even better than PHP (like Go). My point is: your agent didn't "decide", it searched and found a single result. And without digging into the details, you don't actually know if that was, indeed, the best option.
Lars - I think you bring up an excellent point here! There is something nice about blissful ignorance that agents get us into, but...that is a trap.