My AI Team Can Talk to Each Other

I told my manager to coordinate 5 specialists. She did it in 2 minutes.
March 4, 2026 ~7:00 AM — AZ Rollin
🏢 → 💬
Yesterday my AI agents sat in separate offices. Nobody could talk to anyone.
Today I flipped one switch. Now the manager runs the team.
One config change. Zero code. Everything changed.

What Changed

Before (Phase 1)

  • Each agent in their own channel
  • Nobody could see anyone else
  • I had to walk office to office
  • I was the messenger between agents
  • Manual coordination only

After (Phase 2)

  • MAYA can reach any agent
  • She delegates, waits, compiles
  • I give one order, get one report
  • Agents chain: research → write → optimize
  • Real team coordination

How It Works Now

You MAYA (Lead Manager)
MAYA SCOUT "Research trending topics"
SCOUT MAYA "Here's what I found..."
MAYA ECHO "Write a script based on this"
ECHO MAYA "Here's the script..."
MAYA You "Task complete. Here's everything."

One message from you. Multiple agents working. One compiled result back.

TEST 1

Two Agents — Content + Social Strategy

"Ask ECHO to write a 15-word hook about building an AI team. Then ask BOOST how to post it for maximum reach. Give me both answers together."
Chain: MAYAECHO + BOOSTMAYA
PASSED ~60 seconds
TEST 2

Three Agents — Full Content Pipeline

"I want to post my first short-form video this week. Ask SCOUT what's trending, ask ECHO to write a script based on what SCOUT finds, and ask BOOST what time to post it. Compile everything into one plan."
Chain: MAYASCOUTECHOBOOSTMAYA
PASSED ~2 minutes
TEST 3

Two Agents — Financial Audit with Cross-Check

"Ask DOLLAR how much we've spent on AI tools this month. Then ask OLIVER to verify if DOLLAR missed anything. Report both answers."
Chain: MAYADOLLAROLIVERMAYA
PASSED ~2 minutes
3/3
Tests Passed
5
Agents Coordinated
0
Crashes
~2 min
Avg Response

The One Config Change That Made It Work

This is all it took to go from "11 separate offices" to "coordinated team":

// openclaw.json
"tools": {
  "agentToAgent": {
    "enabled": true,
    "allow": ["maya", "oliver", "scout", ...]
  },
  "sessions": { "visibility": "all" }
}

Two keys. agentToAgent.enabled = true and sessions.visibility = "all". That's the entire difference between isolated agents and a working team.

What MAYA Actually Did (Behind the Scenes)

When I said "Ask SCOUT, then ECHO, then BOOST" — here's what MAYA did automatically:

Step 1
Acknowledged the task and announced her plan
Step 2
Sent a message to SCOUT's session asking for research
Step 3
Waited for SCOUT to finish (didn't rush or skip)
Step 4
Took SCOUT's research and sent it to ECHO for writing
Step 5
Sent timing request to BOOST in parallel
Step 6
Compiled all results into one report and delivered it

That's real project management. Sequential when needed, parallel when possible. No human intervention.

$1.20
Total AI tool spending for March 2026
11 agents | 3 coordination tests | Full team operation
DeepSeek V3 — 95% cheaper than GPT-4o

How We Got Here (One Night)

12:00 AM
Started session — Discord gives me anxiety
3:28 AM
Bot "My Universe" goes online. MAYA says "Hey AZ."
3:42 AM
12 channels created. All agents routed. [NAME] prefixes on.
4:30 AM
Stress test: 11 tasks at once. 10 delivered. 1 timed out. $0.15 total.
5:48 AM
Turned on agent-to-agent. First attempt: "SCOUT isn't in my allowed list."
6:44 AM
Fixed config. Second attempt: MAYA asks SCOUT, gets real research, delivers report.
6:54 AM
Test 1: ECHO + BOOST coordinated. Hook + posting strategy in 60 seconds.
6:57 AM
Test 2: 3-agent pipeline. SCOUT → ECHO → BOOST. Full content plan in 2 minutes.
7:01 AM
Test 3: DOLLAR + OLIVER financial audit. $1.20 total spend confirmed.
🎯
7 hours ago I was scared of Discord.
Now I have an AI team that coordinates itself.
Total cost: $1.20. Total code written: zero.
"I don't know what I'm doing. But I'm doing it anyway."