More from the Build-in-Public series
Learn how we structured the team in The 8-Layer AI Agent Stack.
Six weeks.
That's what it cost to build a custom screen to manage our team of AI helpers.
Six weeks of figuring out how to connect the pieces, keep everything updated, handle notifications, and make it usable on a phone.
It looked clean in a demo.
Week three of actually using it in real work, it started breaking.
Not because the build was sloppy.
Because the helpers were still changing. Every time one of them started doing something slightly differently, the screen had to be updated to match. The thing that was supposed to make my life easier was creating more work.
I deleted it.
Then I tried a Slack setup instead.
That lasted eleven days before the same friction came back.
Then one Tuesday morning I was sitting in a parking lot in South Florida with my phone out.
Three tabs open on my phone, trying to figure out why the status screen was showing the wrong state for a task that had already been done.
I closed all three tabs.
Opened Discord.
Typed @Wren in our team channel with a short description of what I needed.
Wren replied with one message. One Google Drive link. The work was in the doc.
I read it. Gave it a thumbs up. Done.
No screen to log into. No form. No complicated setup. Just a message and a Drive link.
That was six months ago.
The custom screen and the Slack setup are both deleted.
Discord is where our ten AI helpers work every day.
This is the post about why.
Why Building a Custom Screen for Your AI Helpers Almost Always Backfires
There's a pattern I've watched repeat across every person I know who's serious about using AI helpers in their work.
The helpers start working. A few tasks get done successfully. Then someone asks the natural question: how do you actually talk to these helpers day to day?
The natural answer sounds like: build a custom screen. A nice-looking app where you type in tasks, see the status, and approve the work. Something that looks like a real product.
The problem is that building that screen is a real software project.
You need the screen itself, a layer to connect it to the helpers, something to track which tasks are in progress, a system to send you notifications, a way to control who sees what, and a version that works on your phone. A good team can build all of that in three to six months.
And almost everyone starts building that screen before the AI helpers themselves are working reliably.
The screen becomes the distraction from the actual hard problem.
I know this because I was in that parking lot on a Tuesday morning staring at a status screen that was showing the wrong information about a task that had already been finished.
The screen I built was adding problems, not solving them.
So here's what I found when I stripped everything away and ran things through Discord instead: the problem was already solved. It just wasn't called an AI helper screen. It was called a group chat.
What Discord Already Does Right (In Plain English)
Discord is not a chat tool you're stretching to do something it wasn't built for.
It's a set of tools for coordinating a team. And those tools are exactly what you need to run a group of AI helpers.
Here are the six things Discord does that would take you months to build yourself.
1. Channels Put Everyone in Their Lane
Every client, every project, every topic gets its own channel in Discord.
Your AI helpers are only added to the channels they need. They can see their channels and nothing else.
That's a real boundary. Built by Discord. Not something you have to code.
Think of it like a building where each floor is a different department. The person on the copy floor can't walk into the research floor without a key card. Discord works the same way, except you set it up by checking boxes, not writing code.
For our ten helpers, this replaced something I was planning to build for months. Wren sees the copy channels. Scout sees the research channels. Sammy sees everything because Sammy is the one who figures out where each task should go. No custom setup. Just checkboxes.
2. Tagging with @ Sends the Task
The @ tag is the simplest way to assign a task that exists.
Type @Wren with a brief description and Wren picks it up. Type @Scout with a question and Scout handles it. Type @Pixel with a concept and Pixel writes the script.
You already know who to tag. The tag is the handoff. There's no routing system to set up, no extra step.
Think of it like tapping a team member on the shoulder and handing them a sticky note. That's the whole interaction.
This matters a lot when you have specialized helpers. @Wren for copy. @Scout for research. @Pixel for video scripts. @Chief for strategy. @Forge for technical builds. You think of the skill you need, you tag the helper who has it.
3. Emoji Reactions Are Your Approval Button
This is the one most people don't think about, even people who use Discord all the time.
When Wren posts a finished draft and I tap ✅ on my phone, that is an approval.
The helpers are set up to watch for reactions on their messages. ✅ means “go ahead.” 🛑 means “hold on, wait for my notes.”
That's a full approval system in two emoji.
No form. No email chain. No status dropdown in a custom app.
I run the whole approval loop from my phone. I read the work, tap the emoji, and the team picks up the next task. The whole thing works without opening a laptop.
Think of it like a thumbs up at the end of a presentation. Everyone in the room knows what it means.
4. Threads Keep Each Task in One Place
Every multi-step task gets a thread in Discord.
The research phase, the first draft, the revision, and the final approval all happen inside one thread, in order.
Scout drops a research summary into the thread. Wren reads it and writes the copy in the same thread. I leave a comment. Wren revises. I tap ✅.
No separate doc to track what's happening. No chat that loses context.
The thread is the task. Start to finish. With full history.
Think of it like a sticky note that never falls off the wall and keeps a record of everything that happened on it. That's a thread.
If a client ever asks “what did the AI helper do with that brief,” the thread is the answer. Every message, every step, every human note. Searchable. Permanent.
5. Getting a Helper Into Discord Takes an Afternoon
Each AI helper gets a simple key that lets them post messages to Discord.
You don't need a developer. You don't need a software team. You create a Discord app, generate a key, and tell the app which channels it can post in.
One afternoon. Not weeks.
And the helper posts under its own name. Wren posts as Wren. Sammy posts as Sammy. When you look at the channel, it looks and feels like a real team communicating. Not a robot outputting text.
Think of it like giving a new employee a keycard to the building and showing them which rooms they have access to. That's what this is.
6. It Works on Your Phone
This one sounds obvious. It isn't.
The entire approval loop for our team runs on my phone. Drive links open right inside Discord mobile. The emoji reactions are one tap. The whole interaction from sending the task to reading the work to approving it works without sitting down at a desk.
Most custom screens fail on mobile quietly. They're built on a big screen, tested on a big screen, and then someone tries to use them from a car or a coffee shop and realizes nothing works right.
Discord's phone app is excellent. It's been that way for years. That's not an accident. That's eleven years of work that you benefit from for free.
The One-Message Rule That Keeps It Clean
The Discord setup is half the system.
The other half is one simple rule we apply to every helper.
When a task is done, post one message. That message has one Google Drive link. The work lives in Drive. The helper never dumps everything into chat.
One message. One link. Done.
Why does this matter?
Three reasons.
First, it keeps the channel readable six months from now. A channel full of raw AI output becomes impossible to navigate within a week. A channel full of labeled Drive links is still clear a year later. Think of the difference between a clean inbox with linked attachments versus an inbox where someone pastes the entire email body three times.
Second, it makes every piece of work easy to find and update. Drive keeps a full history. If I edit the doc after Wren delivers it, both versions exist. Nothing gets buried in the chat.
Third, it creates a clean signal that the task is done. There is no “I think it might be ready” ambiguity. Either there's a Drive link in the channel or there isn't. That's how you know.
Here's what a full campaign kickoff looks like under this system.
A client brief comes in. Wren writes the ad copy and drops a Drive link. Pixel reads the link, writes the video script, drops a Drive link. Chief reads both and drops a strategy doc link. I open three links on my phone and tap ✅ on each.
Complete campaign kickoff. Every piece of work in Drive. Full record in Discord.
No custom screen. No status panel. No complicated connection layer between the helpers and the interface.
Under an hour, start to finish.
What the Team Actually Looks Like
Six cloud helpers run through our Discord every day.
Sammy is the one who figures out where each task should go. When a message comes into the team channel without a tag, Sammy reads it, decides which helper should handle it, and sends it with an explicit tag. Think of Sammy as the front desk person who knows exactly who to transfer your call to.
Wren handles copy. Tag Wren with a brief, get a Drive doc back.
Pixel handles video scripts. Tag Pixel with a concept, get a script back.
Scout handles research. Tag Scout with a question, get a research summary back.
Chief handles strategy. Tag Chief with a campaign goal, get a strategy doc back.
Forge handles builds: landing pages, technical setups, system configurations.
All six are always on. Discord is the place you talk to them. Drive is where the work lands. Emoji reactions are how you tell them what to do next.
For our four local helpers (Sal, CCP, Cleo, Cliff), the setup runs through a different technical layer, but the same one-message rule applies. They post to the same channel. The same thumbs up moves the task forward.
Two different technical setups. One Discord server. One set of rules.
The Real Point
Custom AI helper screens have a problem that goes deeper than the time it takes to build them.
They exist somewhere your team doesn't already live.
A new app is a new tab your team has to remember to open. A new surface with its own rules. One more thing to check on top of everything you're already checking.
Discord is already where your team shows up. Or close enough.
The notification system is there. The phone app is there. The habit of checking it is there.
When your AI helpers post in a channel your team already watches, the updates show up without anyone having to remember to go look. There's no “remember to check the AI helper app” habit to build.
Here's the bigger picture.
The place where you give your AI helpers tasks doesn't have to be something you build.
It just has to be the place where your team already coordinates.
Discord has channels, tagging, approval reactions, and task threads because humans needed those things before AI helpers existed.
The AI helpers are just new team members.
What to Do This Week
If you're in the early stages of setting up AI helpers, here's the fastest path to a working place to talk to them.
Stand up a Discord server. Add your helpers with simple posting keys scoped to the right channels. Decide which helper handles which type of task. Apply the one-message rule: when a task is done, one message, one Drive link.
That is a Friday afternoon.
Then spend the time you saved making better helpers, not a better screen.
The screen question will come up again when your team is running well and you understand the patterns. At that point you'll know exactly what a custom interface needs because you've watched real people use the system. You'll build less, and it will fit better.
For now: build helpers, not screens.
The Mentor Lab is where I share the playbooks behind this team, including the full Discord setup, the one-message rule, and the patterns behind everything I've been building.
It's a private list for coaches, consultants, course creators, and service providers who want early access to the experiments, trainings, and tools coming out of this agency.
If that's you, join here: https://1fb.me/lab-skool
Sam Bell III is the founder of Social Ads Mentor. He runs a real ad agency with a ten-helper AI team and publishes the build-in-public Mentor Lab series for coaches, consultants, course creators, and service providers who are building with AI.
