[untitled]
4 comments
Article ends with "Get in touch with me (LinkedIn, email) if you want to get some details about a product we built aiming at giving control back to the engineers.". Already then.
And execs in SaaS are in the process of killing their business as well.
I have paid an excessive amount of money for trivial bookkeeping software for years, and would have kept on doing if they hadn't destroyed their UI, jacked up rates excessively and forced me to try a slightly cheaper plan that hid that it no longer integrated with tax software, and generally made customers miserable. The proliferation of extra UI elements that hide familiar and fast workflows, and extra AI buttons getting in the way, and excessive advertisements everywhere in the app and by snail mail and email...
If you try to make your customers miserable by extracting every penny from them, they will show zero loyalty, deservedly.
I have paid an excessive amount of money for trivial bookkeeping software for years, and would have kept on doing if they hadn't destroyed their UI, jacked up rates excessively and forced me to try a slightly cheaper plan that hid that it no longer integrated with tax software, and generally made customers miserable. The proliferation of extra UI elements that hide familiar and fast workflows, and extra AI buttons getting in the way, and excessive advertisements everywhere in the app and by snail mail and email...
If you try to make your customers miserable by extracting every penny from them, they will show zero loyalty, deservedly.
This is an ad FYI
> So it skips the root cause analysis you requested.
Stands to reason. Root cause analysis is a process, not a result. With no expectations on what the result should be, the LLM is free to do whatever it wants. What you really need is to ask it to produce failing test case(s). That forces it to do root cause analysis and provide a written explanation of what went wrong, while also providing the feedback mechanism necessary to show that the problem is fixed once you do ask it to proceed with implementation which helps satisfy its "software engineer" predisposition.
LLMs aren't humans that can understand contextual intent like humans can, just fancy compilers that can handle informal languages. You still need to program them like the computers they are.
Stands to reason. Root cause analysis is a process, not a result. With no expectations on what the result should be, the LLM is free to do whatever it wants. What you really need is to ask it to produce failing test case(s). That forces it to do root cause analysis and provide a written explanation of what went wrong, while also providing the feedback mechanism necessary to show that the problem is fixed once you do ask it to proceed with implementation which helps satisfy its "software engineer" predisposition.
LLMs aren't humans that can understand contextual intent like humans can, just fancy compilers that can handle informal languages. You still need to program them like the computers they are.