AI was supposed to make development much more efficient and fun, because now you could choose what to work on and what to outsource to AI agents to implement. Similar to how you manage a team of devs. While we are seeing some examples of utopia I described above… mostly we see devs outsourcing all of the thinking to AI, stamping PRs and then when it breaks… trying a better model to fix it.
I hope many people can relate to what I wrote and also can look at their own SDLC and improve. I went deep into weeds with DSA, and you don’t have to do it with every feature you implement. Just think about the problem, define interfaces and tests and let coding agents implement it. Then use them to iteratively improve your feature to make it more efficient, accurate - better than you could ever make it by hand.
I keep trying to convince companies to implement AI upstream, in the planning stages, which results in clean spec-driven AI SDLC and a huge improvement in communication. Instead, they all want developers to write specs based on their understanding of the task at hand or hook up agents to ticket systems. It's insane that companies want to build from the top but won't look at the foundations.
>AI was supposed to make development much more efficient and fun, because now you could choose what to work on and what to outsource to AI agents to implement. Similar to how you manage a team of devs.
When was ever "managing a team of devs" fun?
Especially for people who liked coding, management was always dreadful. If you went for it, you went for career reasons, not because it was "fun".
I enjoyed the idea of giving boring work to someone else and working on stuff I want myself. But that only worked if I code at night, after I waste my whole day in meetings
AI was supposed to make development much more efficient and fun, because now you could choose what to work on and what to outsource to AI agents to implement. Similar to how you manage a team of devs. While we are seeing some examples of utopia I described above… mostly we see devs outsourcing all of the thinking to AI, stamping PRs and then when it breaks… trying a better model to fix it. I hope many people can relate to what I wrote and also can look at their own SDLC and improve. I went deep into weeds with DSA, and you don’t have to do it with every feature you implement. Just think about the problem, define interfaces and tests and let coding agents implement it. Then use them to iteratively improve your feature to make it more efficient, accurate - better than you could ever make it by hand.
I keep trying to convince companies to implement AI upstream, in the planning stages, which results in clean spec-driven AI SDLC and a huge improvement in communication. Instead, they all want developers to write specs based on their understanding of the task at hand or hook up agents to ticket systems. It's insane that companies want to build from the top but won't look at the foundations.
> clean spec-driven AI SDLC
Sounds like waterfall development with new drapes, we know how well waterfall works
software is iterative and we know that you only discover things once you actually get into it
>AI was supposed to make development much more efficient and fun, because now you could choose what to work on and what to outsource to AI agents to implement. Similar to how you manage a team of devs.
When was ever "managing a team of devs" fun?
Especially for people who liked coding, management was always dreadful. If you went for it, you went for career reasons, not because it was "fun".
I enjoyed the idea of giving boring work to someone else and working on stuff I want myself. But that only worked if I code at night, after I waste my whole day in meetings