Software architect with strong opinions | Offering 15+ years of knowledge to Tech professionals | I share insights and real world experiences in the IT business
The time has come.
The middle managers at my client have seen what AI can do for deva and now they are going all the in.
This per se is a good thing. As always in these cases the implementation is suboptimal.
They created a dedicated team trying to implement all at once and in the classic ultra bloated corporate version.
In matter of a week I saw Pull requests trying to add 20 skill and 50 reference files of 1000+ lines each.
When I questioned the answer was “a guideline for the coding agents os better than no guideline.”
This is not how LLM works, this is not how a coding agent should learn about your guidelines.
AI people… keep your skills nice and small please. The more the merrier doesn’t work with LLM context.
@mattpocockuk I had a discussion last werk aboit this. 10 pages long skills can only promote mixed answer and hallucinations. Everyone is using them as a cheap RAG but with mixed results
RESTful APIs may be dead soon. Instead, web services may expose a single POST entry point for a prompt. Internally, an AI agent may decide how to interpret it and what to do with the data and the database.