What I find really disturbing is the impossibility to catch up with the development - which agent system to use, which model, what are the right models for what tasks?
I do not fear that some agents will pollute my repos with their PR. In opposite, I suppose that we will end at a point where for each question, task, or problem, one will find many (AI-coded) solutions, making it impossible to choose a right, solid, reliable one. I recently thought about having a database of tools per task so that a comparison would be possible. But the maintenance costs of something like this are enormous when including benchmarks, comparisons, etc. on different qualities.
I do not fear that some agents will pollute my repos with their PR. In opposite, I suppose that we will end at a point where for each question, task, or problem, one will find many (AI-coded) solutions, making it impossible to choose a right, solid, reliable one. I recently thought about having a database of tools per task so that a comparison would be possible. But the maintenance costs of something like this are enormous when including benchmarks, comparisons, etc. on different qualities.