I requested of Google search "generic word for tools like openclaw" and it presented a Gemini result at top of page (per usual) that reads "...AI agents..." which result pane then disappeared after about 3 seconds replaced with the Gemini logo and the message "Can't generate AI overview right now. Try again later"

Unfortunately cute, because as "AI agents" are being virtually lashed together with scripts and baling wire (the industry seems to have bet the farm on an AGI to rule them all, and now is scrambling for yet-another-framework to make AI units computationally useful) how are engineers supposed to think about program correctness when the units of computation they employ may produce effectively random results to any input, and different results whenever the agent is updated?

As constantly changing agents are promoted as a great boon in the AI industry, Does the discipline of software "engineering" effectively lose all meaning when the computational unit being marshaled are fundamentally unpredictable with respect to their inputs? "Debugging" as a term returns to its original meaning of insects inside a machine causing short-circuits.

Who can honestly endorse the deployment of life-sustaining infrastructure built from such mechanisms?