I built a team of 11 AI agents that live in Discord. A researcher, a writer, a finance guy, a social media strategist — the whole company. My main AI brain (Claude Code) can send them tasks through Discord webhooks.
Earlier tonight, I dispatched 6 tasks to the team. Scripts, posting strategies, cost breakdowns, retention scoring. They all delivered. The system worked perfectly.
Or so I thought.
Two-way communication. Send tasks, receive results. Right?
At around 4 AM, I casually asked Claude:
Wait. WHAT?
My main brain — the one that controls my entire system — couldn't hear a single thing the team was saying.
MAYA had finished a complete analysis of a YouTube channel I asked about. Assigned SCOUT to research it. ECHO built a content plan. They tagged Claude Code when they were done. The message sat there for hours.
Nobody was listening.
Sending worked. Receiving was completely broken.
Discord webhooks are one-way only. They can send messages but can't read them.
I had been using a megaphone to shout at the team.
But I had no ears.
When I set up the communication between Claude Code and Discord, I used webhooks. Webhooks are like a one-way loudspeaker — you can broadcast messages, but you can't hear anyone talking back.
Think of it this way:
Webhooks = a megaphone. You can shout through it. But you can't listen through a megaphone.
The agents were responding. MAYA was tagging "@Claude Code" when tasks were done. SCOUT was delivering research. ECHO was writing scripts. But all those messages were going into Discord and just sitting there, waiting to be read by an AI that had no way to read them.
I only found out because I manually asked: "Did you see MAYA's message?"
If I hadn't asked, we would never have known.
discord-inbox.md. Messages mentioning Claude get flagged.Full duplex. Send tasks. Hear results. No human relay needed.
This message sat unread for 7 hours.
Claude Code had no way to see it. AZ didn't know to check. The work was done — but the loop was broken.
If you're building any kind of AI system — agents, bots, automation — test the full loop. Not just "can I send a message?" but "does the response actually come back to where it needs to go?"
We had 11 agents. Webhooks. Discord channels. A manager that delegates tasks. Agent-to-agent communication. A stress test that passed. It all looked perfect.
But the main brain was deaf.
And we only found out because at 4 AM, one casual question exposed the entire gap.
Now it's fixed. 12 channels monitored. 30-second refresh. Every message captured. Every @mention flagged. The brain has ears.
"I don't know what I'm doing. But I'm fixing it anyway."