How Do You Build An AI Agent? Be BRIEF
A simple framework to help you build your first AI agent. It’s not a new technical task to master. Just think of it as a new employee to onboard. What are the 5 things someone would need to do the job right? That’s BRIEF.
An employee walking their new AI Agent through all the company’s TLAs*. Created by author with Google Gemini.
Lately I’ve noticed that the topic of AI agents have crossed from the technical community into my LinkedIn feeds and boardrooms. People are being forced to get up to speed, either because they want to get ahead of the curve or they have been mandated to “figure it out”. And there are plenty in that second camp based on my personal experience.
I feel bad for the “figure it out” crowd. They have to find the time to add one more thing to their plate, learning a new technology and methodology on top of everything else. And since the tools constantly change, by the time they learn it that version will already be obsolete.
Fortunately, I don’t see understanding and creating AI agents as a technical skill to develop. When I boil it down, it’s simply a new version of a classic management challenge. How do you get the new guy up to speed? You know how when you get a new hire on the team, you have to take the time to explain what exactly you need them to do so that they can actually be good at their job? Building an agent is essentially the same thing. What do you do to set that person and the team up for success? The advice doesn’t change, whether it’s a new hire or a new AI: be BRIEF.
The BRIEF framework below is purposefully, extremely non-technical. But to give a tiny bit of the technical perspective, I’ll give you my definition of what an AI agent is. An AI agent is basically a “mini-AI” tool that is focused on what you say and follows the specific instructions you personally put in place. Think of it as a chatbot that follows your directions that doesn’t ask you to repeat itself and doesn’t waste time (and tokens) going on random wild goose chases across the internet.
Let’s say you want to create an agent that writes a draft of your weekly client update emails. You don’t need an agent that references everything that’s ever been written. You need it to simply reference the key notes and materials stored in the CRM and shared drive to put something together that makes sense. This isn’t Shakespeare, you just need a good first draft.
You’re building something that is fast, reliable, and repeatable. If you hired someone, you’d do the same. Explain the job, give them the inputs, and tell them what a good output looks like. It’s as simple as that. So I tell people to answer these 5 questions and document the answers to every single one.
The BRIEF Framework for Building AI Agents (and yes, this also works for hiring real-life humans, too)
Background: What does this agent need to know about your business, your customers, your processes?This is the context. It can be multiple documents that are always referenced. Most people skip this entirely and wonder why the agent produces mediocre output.
Role: What is this agent’s job for this specific task? Don’t be generic (“you help with marketing”) when you can be specific (“you are the first-draft writer for weekly client update emails”). Don’t have an agent that wears multiple hats and does different jobs. This one agent has this one job. You can always create more.
Instructions: What are the rules that must always be followed?This is the employee handbook and guardrails. Don’t trust AI to send something directly to the client without a human review? That’s a good rule. Tell it never to do that. Specify the language it can and cannot use. Decide what requires human input and what can be delegated to AI. It will follow the rules if you set them.
Expectations: What does a great output look like?Give it examples of acceptable and unacceptable work products. This is critical for quality control. Surprisingly, most people miss this. They don’t realize they have more control over the AI outputs beyond just the prompts they give it. Just like you would with a new employee, articulate what a job done right looks like and set the standard to follow.
Feedback: How do you improve it over time?This actually goes two ways. The first is that you review AI’s work and if you see consistent errors, you give that feedback (ask AI why it keeps doing it) and update your BRIEF documents as necessary. The second version is to tell the AI it should ask you questions when it is unclear or needs more detail. AI is built to give you answers. If it has incomplete information, it will guess. So make sure the feedback loop goes both ways.
That’s it. Nothing you haven’t done before. You don’t need to be an AI expert to follow this method. You just need to know what is required to do the job right.
You likely have a lot of this documented already, either in one place or multiple. The easiest way to do it is to write it down, pull from existing process guides, or simply record yourself talking as you pretend you are onboarding a new hire for this specific task. Take the documents or transcript, feed it into the chatbot of your choice, and have it give you the first BRIEF draft. Give it a final review and then there you go. You have what you need to build your first AI agent.
P.S. I spent 20 years in Corporate America. I love me a good acronym. My favorite corporate acronym of all time? TLA. It stands for Three Letter Acronym. As in “we have a lot of TLA’s here”. Consider this your corporate dad joke of the day.
