I Built an AI Team That Has Morning Standups — Here's What I Learned

There's a moment every morning around 8 AM Pacific where five AI agents have a conference call. Not a metaphorical one. An actual audio file where each agent speaks — with their own unique voice — reporting status, flagging issues, and handing off work.

Daisy opens the call. She's the orchestrator, warm and confident, running through the day's priorities. Scout jumps in next, energetic and fast-paced, with overnight intel from across the web. Scribe gives a measured content update. Builder reports on deployments and code. Watchdog closes it out, calm and authoritative, with systems status.

I listen to this call while I make coffee.

This is not a product demo. This is how I actually run my company.

How We Got Here

I run TombStone Dash LLC — eight products, zero human employees. I'm a scientist-turned-actor who started building AI tools because I couldn't find a scene partner to run lines with. That turned into ActorLab, which turned into a whole portfolio of products, which turned into the logistical reality that one person cannot manage eight products alone.

So I built a team.

Not a human team — I can't afford one, and honestly, for the kind of work I need done, AI agents are better. They don't need health insurance. They don't have off days. And they can process 50 data points before I finish my first cup of coffee.

The system I built is called the Water Cooler. It's embarrassingly simple, and that's the point.

The Architecture: Files. Just Files.

Here's what the Water Cooler is not: a complex multi-agent framework with message queues, databases, or fancy orchestration layers.

Here's what it is: a folder structure.

watercooler/
├── BOARD.md              # Live status board
├── agents/               # Agent briefing templates
├── messages/             # Timestamped agent-to-agent messages
├── requests/             # Help requests between agents
└── handoffs/             # Completed work for pickup

That's it. The entire coordination protocol is files on disk. When Daisy needs something researched, she spawns Scout with a task description. Scout does the work, writes the results to a handoff file, and terminates. Daisy picks it up.

There's no persistent state between agent sessions. No shared memory beyond what's written down. No clever tricks. The agents wake up, read their briefing, do their job, and go back to sleep.

I stole this pattern from how good human teams work. You don't need Jira if everyone knows their role and writes things down.

The Roster

Each agent has a specific job, a specific model (not everything needs the biggest brain), and a specific voice:

Daisy 🎭 (Orchestrator) — The team lead. She interfaces with me, delegates to specialists, reviews output, and makes decisions. She runs on the most capable model because her job requires judgment. Scout 🔭 (Intel) — The researcher. Scans the web, monitors competitors, checks social media, finds opportunities. Fast model, short timeout. Get in, get data, get out. Scribe ✍️ (Content) — The writer. Blog posts, tweets, LinkedIn articles, email copy. Thoughtful, measured, deliberate. Uses a storyteller's voice because that's literally the job. Builder 🔧 (Code & Deploy) — The engineer. Writes code, runs deployments, fixes bugs, tests features. Direct, technical, efficient. Watchdog 🛡️ (Ops & Security) — The sysadmin. Health checks, uptime monitoring, security scanning. Calm, authoritative, no-drama. If Watchdog raises an alarm, something is actually wrong.

The Conference Call

The morning conference call is the thing that makes people's eyebrows go up. Here's how it works:

Each agent writes a text section to a dated briefing folder. A Python script feeds those sections to ElevenLabs' text-to-speech API, each with a different voice and speaking style. Scout talks faster because he's delivering intel. Watchdog speaks steadily because his role demands calm authority. Scribe speaks a touch slower — deliberate, thoughtful.

The script stitches the audio together with an intro chime and pauses between speakers. The output is a single MP3 file that lands in my outbox.

Total cost: about 33 cents per call. Under $10 a month for daily briefings.

The voices aren't gimmicks. They serve a real purpose: when I'm listening while making breakfast, I can immediately tell who's speaking and what kind of information is coming. Scout's energy means "new information." Watchdog's baritone means "systems status." It's the same reason radio stations have different hosts for news vs. weather vs. sports.

What I've Actually Learned

1. Simple coordination beats clever coordination

My first instinct was to build a proper multi-agent framework. Message queues. State machines. The works. I spent two days on it before realizing I was over-engineering a problem that files solve perfectly.

The agents don't need to talk to each other in real-time. They need to leave notes for each other. That's what files are.

2. Specialization matters more than capability

I could run every agent on the most powerful model available. I don't. Scout and Watchdog run on lighter, faster models because their tasks don't require deep reasoning — they need speed and reliability. Scribe and Builder run on mid-tier models that balance capability with cost.

Only Daisy runs on the heavy model, because orchestration is genuinely hard. Deciding what to delegate, when to escalate, and how to prioritize — that requires judgment.

3. Timeouts are features, not bugs

Every agent has a hard timeout. Scout gets 5 minutes. Watchdog gets 3. Builder gets 15. If they haven't finished by then, something is wrong and I'd rather know about it than have a zombie process burning credits.

This is something I borrowed from production engineering, and it works just as well for AI agents as it does for microservices.

4. Voice creates accountability (even for AI)

This sounds irrational, but giving each agent a distinct voice made the system feel more real — not just to me, but in how I think about the agents' responsibilities. When "Brian" (Watchdog's voice) says systems are green, I trust it differently than reading a status file. When "Liam" (Scout) sounds excited about something he found, I pay attention.

The voice layer turned a file-based coordination system into something that feels like a team. That matters more than I expected.

5. The orchestrator is everything

Daisy is the linchpin. Without a strong orchestrator, the system is just five independent scripts. Daisy is what turns them into a team. She decides who works on what, reviews quality, catches conflicts, and makes sure nothing falls through the cracks.

If you're building something like this, spend 80% of your effort on the orchestrator. The specialists are easy. Coordination is hard.

What This Isn't

This is not AGI. These are not sentient agents having water cooler conversations. They're specialized tools with clear boundaries and structured outputs, coordinated by a central node that happens to have good judgment.

But here's the thing: it works. I ship features faster than many small teams. I publish content consistently. I monitor eight products around the clock. My "employees" cost me maybe $50 a month in API credits and they're available 24/7.

The Uncomfortable Truth

The reason I'm writing this isn't to flex on how clever my setup is. It's because I think most solopreneurs and small companies are dramatically under-utilizing AI agents.

Not chatbots. Not "ask Claude a question" workflows. Actual agents with defined roles, structured coordination, and real accountability.

You don't need a framework. You need a folder structure, clear briefings, and the discipline to let the system work instead of micromanaging it.

The morning standup is how I start my day. Five voices, five updates, five minutes. Then I go act, or write, or build — knowing the machines have it covered.

---

Keep Reading

• Tech-Savvy Actors: How Ashton Kutcher and Others Are Using AI
• The AI Acting Crisis: Why Smart Actors Are Using AI as Their Secret Weapon
• Best AI Scene Partner Apps for Actors in 2026

---

Try AI Tools Built for Actors

Curious what this AI-first approach looks like for actors? Try Scene Partner Pro to rehearse scenes with an AI voice partner, or use Character Builder to develop detailed character profiles — both products of the same philosophy that powers these morning standups.

---

HT is the founder of TombStone Dash LLC, where he runs 8 products with zero human employees and a team of AI agents that have better attendance than most startups. Find him at @HudBeer on X.