Articles·AI agents
“I needed the DoBetter Coach to own my blind spots.” Why a custom AI agent beats a general chatbot.
A custom AI agent is an AI assistant built for one job, with its own written rules, its own memory, and its own place to put finished work. It beats a general chatbot because it already knows the job before you type a word. After 17 years of selling health insurance, I stepped into a business I’d never run, in a field so new that most people I could ask had never done it either. I needed the DoBetter Coach to own my blind spots.
Why doesn’t a general chatbot do the job?
A general chatbot doesn’t do the job because it knows a little about everything and nothing about your agency. Every session, you explain the task, your rules, your voice, and what you decided last week. Then it does a decent version of what you asked, in its own style, and forgets it all by tomorrow.
That’s fine for a one time question. It’s slow for the work you do every week: the same reports, the same emails, the same decisions against the same rules.
What do most agencies try first?
Most agencies try one general assistant for everything, or a custom GPT with instructions and a few uploaded files. A custom GPT is a real step up. It holds a standing set of instructions and answers from your documents.
Where it runs out is the work after the answer. A custom GPT answers questions inside a chat window. It doesn’t keep a running memory of what you’ve decided, hand work to another agent, or put a finished file where your team will find it.
Wanna fix it?
A general chatbot that doesn’t know your work gets fixed with custom AI agents: one agent per job, each with written rules, its own memory, and a shared memory underneath so they all know what you’ve decided. Here are the ones I run on Claude.
- The DoBetter Coach. My business coach for pricing, the offer, positioning, and go to market. It works from 9 standing research files and a dated notes file it updates every session. Its tone rule: the friend who happens to have built a business.
- The video editor. It cuts my talking head videos from word level transcripts, builds graphics from two locked design systems, and checks every video against a written list of AI tells to keep out.
- The writers room. It writes these articles and my comedy scripts from a voice brief, a banned phrase list, a log of what’s already been made, and a bank of unused jokes.
- The CRM agent. It owns my CRM: the pipeline, the reply drafts, the meeting briefs.
- Finn. A benefits chatbot loaded with about 300 emails of real questions and answers plus all of the agency’s plans. It sat on every landing page I built.
And one for fun: an advocate agent, a devil’s advocate in reverse. Its only rule is that whatever I believe is true, it has to build the argument for it. No matter if I’m calling the sky pink, it must advocate for what I believe.
What does it look like once it’s running?
Once custom agents are running, you walk into the right office and start working. You don’t explain the job. The coach already has the pricing notes. The writers room already knows the banned list. The video editor already has the design system.
Each agent saves finished work in its own folder and writes decisions back to the shared memory, so the next agent knows what happened. When one gets something wrong, I add a rule to its file.
Can you do this yourself?
Yes, for your first agent. Pick one job you do every week. Write a one page rules file: what the job is, what good looks like, what’s banned, and where the finished work goes. Load it into a Claude project or a custom GPT with your best three examples. Use it for a month and add a rule every time it misses.
A single agent stops being enough when several agents need the same facts, when one needs to hand work to another, or when the work has to land in your CRM or your files. One job, write a good rules file. Five jobs that share clients and decisions, build the agents.
Quick answers
- What is a custom AI agent?
- An AI assistant built for one job, with written rules, its own memory, and a set place for its finished work.
- Is a custom GPT the same as a custom AI agent?
- A custom GPT is a good start: standing instructions plus your documents. A custom agent adds a running memory, handoffs to other agents, and work saved outside the chat.
- What should a small agency’s first custom agent do?
- The weekly task with the clearest rules and the most repetition.
- What is AI agent development?
- Writing an agent’s rules, giving it the documents and memory it needs, and connecting it to where the work lives.
- Has DoBetter built custom agents?
- Yes. Chris Cordon runs his business on custom agents for coaching, video editing, writing, his CRM, and benefits questions.
Next step
I’m Chris Cordon, and I build systems like this for benefits agencies at DoBetter (dobetter.plus). The intro call is free and takes 30 minutes: dobetter.plus/talk.
+Talk to Chris