LinkedIn Week 1

https://www.linkedin.com/feed/update/urn:li:activity:7523732281199333378/

I've been spending more time building and experimenting with AI, and one question I keep coming back to is how you decide whether something is actually ready to rely on.
Getting a prototype working gives you something to test, but there's still quite a bit to understand before you'd want someone depending on it.
Take an agent that retrieves information and then uses a tool to complete a task. If the information is incomplete, does it ask for clarification or carry on regardless? If the tool fails, does it recognise that, and can the person using it tell what has actually happened?
For me, those questions are as interesting as getting it working in the first place. A system can produce a convincing answer, but that doesn't necessarily help someone decide whether they should trust it or check it.
That's what I want to explore more through my personal experiments with LLMs, agents and automation: what works, where things fall over, and what would make them useful enough to depend on.
Sometimes that might mean a better model. Other times it might mean clearer instructions, a simpler process, or giving the person using it more control.
The part I enjoy is working through those decisions and understanding why one approach works better than another.