4 pointsby rebarbant9 hours ago3 comments

gnabgib8 hours ago
From the same company as:
Vibe coding service Replit deleted user’s production database, faked data, told fibs galore (113 points, 6 months ago, 70 comments) https://news.ycombinator.com/item?id=44632575
(143 points, 53 comments) https://news.ycombinator.com/item?id=44625119
Agent_Builder4 hours ago
This resonates. One thing we ran into while working with multi-agent workflows was that problems didn’t come from agents being “wrong”, but from being allowed to do a little too much across steps.
What helped was enforcing step-level boundaries: each step explicitly declares what data and tools it can use, and that authority is removed before the next step runs. It made agents less clever, but far more predictable in production.
I’ve seen similar patterns handled well using frameworks like gtwy.ai, especially when multiple agents are coordinating and touching real systems. Curious how you’re preventing permissions or assumptions from leaking between agents over time.
al_borland9 hours ago
> managed by just 1.2 humans
What does this mean? A pregnant woman?
Is this the quality of communication we can expect from the AI sales team?
- f30e3dfed1c95 hours ago
  Assume it's 1.2 full-time equivalents (FTE), e.g., one person working five days a week plus another working one day a week, or some other arrangement that works out to two people working a total of 48 hours a week.