with LLMs in a few years software would be a low level thingy that people used to spend hours writing and debugging. like assembly back in the day. software is becoming more and more cheaper to put in place. craftsmanship is in the product and value not in the software itself. we are in an era: adapt or get out ...
except LLMs are completely useless for solving complex problems and proper engineering fundamentals for large projects and that fact has not changed one bit since 2022. They're better and better at CRUD app boilerplate that doesn't go into any detail at all whatsoever because they are using the training data of 1000s of repositories for that, but there simply is not enough training data remotely for anything novel and complicated
Switched to an AI team 8 months ago. Ended up doing infra and api integration for llm tuning which is a nice way to start ramping up to the field. With every integration learned new things: Lora/pets, spin up servers to host the models for inference/tuning. It's a lot of learn on the spot compared to concepts from SWE life.
I had a very good experience using safe 4.x
The whole department participated.
Edge Software
Cloud platform
Application team
It allowed the 3 teams to have better communication, things were written, we knew what we are building and how/who is using it.
Met with stakeholders regularly for feedback and better understanding of the business in general, the strategy and why we need to land this feature either to differentiate or because it is priority 0 asked by many customers.
As a team member safe allowed me to understand why what I am working on is important for the business and how it brings value.
I wonder if it is a people problem that we are blaming safe for or an actual process problem?
Some problem might be it's both a people and a process problem. There may be little to nothing about it that is objectively, technically "wrong". But maybe it doesn't work for 90-95% of people.
It sounds like most of what you listed is something that could be accomplished with a good scrum team that has a well-defined, prioritized, and pruned backlog (as necessary.... :))
