Kelly Chan

Let's do an honest agent roast. If you've tried using an AI agent in real work, tell us about a time it didn't work.
What were you trying to get done? What did the agent do instead? What did you have to redo or clean up? Stories from sales, operations, research, support, coding, or anywhere else are welcome.
Maybe it lost context between tools. Maybe it lacked permission, made something up with confidence, or broke in a completely different way. No polished demos, no pitches, and no need to have a fix. I'm curious which failures keep showing up when agents leave the demo and enter daily work.
What's your most memorable agent fail?

Hi, I'm Kelly from Busabase. We're building an open-source database and workspace where the records, docs, skills, and apps a team chooses to keep can be reused by people and connected AI agents.
One problem we keep running into: an agent finishes useful work, but the result stays buried in a chat. The next person or agent has to reconstruct the project background and decisions before moving forward. In Busabase, that work can live in a shared workspace, with access controlled by permissions and review when needed.
How does your team handle this today? Where do you keep the project context that multiple agents need, and what gets lost most often when you move between tools like Claude Code and Codex? I'd love to hear what's working for you and what still breaks.

Hey PH, I'm Kelly, founder of Buda, launching today:
https://www.producthunt.com/prod... Earlier this year, I handed 90% of our company's daily work to AI agents. Two weeks in, 40% of my team didn't know what to put in their weekly update. Not because they weren't working, but because agents had quietly cleared their to-do lists. They stopped doing the work and started managing the agents who did. Support, ops, design: all became AI managers overnight. The infrastructure was chaos though: 8 Linux machines, 2 Mac Minis, 30+ chat channels. So we rebuilt it into a product: Buda. Each Claw agent gets its own computer, browser, and disk. You open a website, click hire, and your agent starts working. No Mac Mini. No setup. No model config. Happy to talk about agent management, token costs, what actually works vs. hype, building in public, or anything else. AMA.