@tausif_khan099 If agents start doing the buying, the page has to win on specs, price and the return policy, since that is what the agent will compare.
@tshokama Good habit, Glory. A running list of misses shows where the tool is safe to trust, and a product team that keeps one can brief the next model on the exact work it still cannot do.
@ibx1112599 Separate owners is the smart part. Procurement already trusts its own contracts people more than a general helper, so an agent that team owns starts with that trust. The combined answer is where it gets tricky, since that seam has no owner yet.
@Apoorvajosh Clear example, Apoorva. Cutting the tool that made you 20% faster looks like a saving on paper, and the EOD deadline stays exactly where it was.
@yantowid1 Sharp way to put it, Yanto. An agent that held the 2am stop would have saved the trade and skipped the lesson that taught you to size small.
@Galo2027 So the safest agent for my Citi account is whichever one Citi built. Terms that only bite outside agents work like an ad for the bank's own, and picking an assistant for my money starts to look a lot like picking a bank.
@kanesimms Great framing, Kane. The key word in United's rule is unauthorised, so the door stays open for agents it signs off on. That turns the question into who gets approved, and an airline choosing its favorite AI agents looks close to how it once chose its favorite travel agents.
@hivinz_ Love this design. Putting where the agent waited for a human right next to how long each step took makes it easy to tell a slow model from a stuck handoff, and that is the first thing a support lead will ask about a late ticket.
@zuess05 Part of it is that a human bug comes with an author who can explain what they meant. An AI bug comes with a confident commit message and an author that does not remember writing it.
@s_v_g@Microsoft So the buyer now writes the job description for each agent. Microsoft gave agents names, machines and budgets at Ignite, yet who answers when one gets something wrong still sits with the customer, and that is an HR question as much as an IT one.
@CodeWithSravya Love that you framed it from the agent's side, Sravya. Asking what an agent fails to find on a site is a sharper question than asking how the page looks, and it is the one product teams will have to answer next.
The part that goes missing in solo agent work is the why. A teammate sees the final diff, while the hour of back and forth with the agent that explains each choice stays in one person's chat window. Making that session something the team can read and comment on feels like a natural first step toward multiplayer.
@beginnersblog1 Love this. Five agents look great in a demo and turn into five handoffs that can break once real orders come in. A plain workflow with one agent at the step that needs judgment seems to win for now.
@KobinFlow The unsettling part is how casual it sounds, like a helpful friend dropping your bank balance into a random chat. An agent that reads one app and talks in another needs a hard rule for what can cross between them, and right now that rule seems to live in the model's judgment.
@zaryanbliss Great breakdown. Loading only the tools an agent needs sounds like a small change, yet a long tool list is exactly where agents start picking the wrong one.