I'm not dead, I'm just resting! Loves: holy grails, shrubberies, spanish inquisition and arguing for the sake of it. Obviously sharing only private opinions.
This productivity jump can come up with 9 agents being able to have a baby in one month…
Or fact my car can go 250 km/h - while in reality my speed on average tops 60. (do I need faster and more expensive car?)
Business-wise ignoring fact - companies (and see here - people who work there) can accomodate finite amount of changes or applications. There is always bottleneck somewhere, which makes this theoretical boost not only useless, but also dangerous (affecting quality of the whole process).
Not sure if this is sandboxing case… If I wrote piece of code, it would follow strictly hard constraints of do / don’t when called via API. Testing this in sandbox / test gives certainity for prod environment (assuming code / design is right and supported edge cases).
Here it sounded more like agent decided to ignore constraints (with very broad access) - behaving bit like end user, who knew shouldn’t but still did take wrong action.
Backup policy is of course different story, can’t defend this one.
@omphalus_tech@atmoio Unless your resources are trimmed down by someone else who believed you don’t have to. Leaving this as the only option due to self-imposed constraints.
10 years ago, if I would write a script which does analyze unstructured data, clean them, apply segmentation, propose next steps based on priority (referencing to defined business rules, which are hard constraints) and pushes notifications / action items to specified step owner / end user for approval… I would call it “script which does analyze…”.
Today everything is called AI agent…
@nuhre_@wesleytypes I’m starting to be thankful it was that bad at the start. Made me skip on even more pain later it seems. Solid story with tough choices and strong non ‘black and white’ characters, ruined for what exactly…