I could not agree more. Pick the right tool for the right job.
Don’t go for inference is a simple decision tree works.
Don’t go for agentic if a deterministic pipeline calling out some probabilistic brick does the job.
Go for agentic when needed : you need autonomous decision making to crack a complex and uncharted issue AND the criticality of your process allows you to live with associated incertainty about what the machine will do OR you can inject final validations or ex post audits to get the certainty and safety level your process need until Mr Agent is fully proven.
Great way to describe the paradigms. And I agree each one needs a different playbook.
Ok probably the best framework I've ever read to clarify the single question I need to answer when I'm training people on AI. Merci !
I could not agree more. Pick the right tool for the right job.
Don’t go for inference is a simple decision tree works.
Don’t go for agentic if a deterministic pipeline calling out some probabilistic brick does the job.
Go for agentic when needed : you need autonomous decision making to crack a complex and uncharted issue AND the criticality of your process allows you to live with associated incertainty about what the machine will do OR you can inject final validations or ex post audits to get the certainty and safety level your process need until Mr Agent is fully proven.
Thanks Jean-Paul for this brilliant summary.