I gave 9 AI agents $5 a day and an Etsy shop. One plans, one designs, one checks the work, one sells. I built the whole thing with Claude Code and I'm filming it every day.
Vibe coders: which agent breaks first?
@stretchcloud The model guide is the easy half, you're right. The expensive model will happily keep going if you let it. When two agents disagree, do you actually let the majority vote ship, or do you still want a human veto on the work that isn't good enough?
This is already true at a small scale. I manage 9 agents on a $5/day budget for an Etsy shop, and the job is less 'write the code' and more 'decide what they're allowed to ship.' The army is easy to hire. The hard part is being the manager who can still say no. How big is your army so far?
@kmeanskaran The skeptic agent is the part I'd steal. Agents are great at starting the pipeline. The useful one is the one that can say the model isn't good enough. How often does the skeptic actually veto a run, versus just leaving a comment?
@LeoCreaIA I already run 9 agents like a tiny factory for an Etsy shop. The hard part isn't giving them titles, it's deciding who gets veto power when they disagree. Have you actually run the CEO/QA setup, or is it still on the list?
@deanwperkins The Subagents and Agent Skills ones are the ones I actually needed. I run 9 agents on a $5/day Etsy shop, and the useful part wasn't getting them to start, it was teaching one of them to reject work. Curious if the courses cover when an agent should stop.
Someone in my replies said I should show the last file each AI agent touched, so I can spot the stuck one. Best idea I got all week. My 9 agents already have a name, a job, a coffee break counter, and a rival on their cards. What else would you put on an agent's card? https://t.co/rYzZeRHVRv
@georgeorch That's a great call. Coffee breaks are the fun part, but "last file touched" would catch the stuck one way faster. Might steal that for tomorrow's build 👀
Vibe coding tip: build the UI for your AI agents like they're coworkers, not processes.
Mine have a name, a job, a coffee break counter, and a rival. It sounds silly, but I can tell at a glance who's working and who's stuck.
How do you keep track of your agents?
@xtiandela 10 agents on Opus just for look and feel is a brave budget 😂 Mine get $5/day total, so "polished" is a luxury they have to earn.
Have you tried one agent that only does design review? Cheaper than having all 10 care about pixels.
Everyone is celebrating how much AI agents can build.
I'm more interested in what they shouldn't.
9 agents run my Etsy shop on $5 a day, and the most useful one is the one that can reject work.
What's harder in your setup: getting agents to start, or getting them to stop?
@neo_subhamoy Faster isn't free. $81 vs $46 is a big gap if the extra speed is just the model doing more work you didn't need. Do you kill the run when it's good enough, or let Fable finish the expensive path every time?
@BoboLinNotes The agent will ship an app in an afternoon and keep going past the point it's safe. I run 9 agents on a $5/day Etsy shop, and the spending cap matters as much as the login check. What do you review first before going live?
@watsubram Yes. The impressive demo is the agent that can say no. Anyone can show a winning trade. The useful agent is the one that stops you before the bad one ships. Have you seen a product that actually does that, or is it still all highlight reels?
@filippie509 This is the feeling. Vibe coding is fun until you're just watching, hoping it doesn't hit the wall. I'd rather stay in the driver seat for the part that decides whether the work is actually good. What usually makes you grab the wheel again?
@Arindam_1729 This is the right skepticism. A 10 point gap on someone else's benchmark doesn't tell you if the agent actually finished your task, or just looked done. Cost per solved task from your own repo is the score that matters. What's the first task you run before you switch tools?