If an agent is a digital worker, it helps to know what it is made of. Take one apart and you find a small number of distinct parts, each doing a specific job, working together to make something that can actually perform work.
There are five. Intelligence, knowledge, memory, skills, and awareness. Each one matters, and an agent that is missing any of them is crippled in a predictable way. Understanding the five parts gives you a clear mental model of what an agent is, why it succeeds or fails, and what you are actually building when you build one.
This article takes an agent apart, part by part, so you can see what makes it work.
An Agent Is Made Of Parts
It is tempting to think of an agent as one thing, a black box that does work. But the box has parts, and seeing them is what lets you reason about it.
The five parts are intelligence, knowledge, memory, skills, and awareness. Intelligence is the reasoning. Knowledge is what it knows about your business. Memory is what it remembers over time. Skills are what it can actually do. Awareness is its sense of its own job and limits. Together, these five turn a raw model into a worker. Let us look at each, and at why each one is necessary.
Intelligence: The Reasoning
The first part is intelligence. This is the raw reasoning ability, supplied by the underlying AI model.
Intelligence is what lets the agent understand a request, think through what needs to happen, and figure out how to respond. It is the engine of reasoning at the center of the agent. It is the general reasoning ability of the model. It is powerful, general, and, by itself, ignorant of your business and incapable of action.
Intelligence alone is not an agent. It is the thinking part, and thinking with nothing to think about and no way to act is not work. The other four parts are what turn intelligence into something that can actually do a job.
Knowledge: What It Knows About Your Business
The second part is knowledge. This is everything the agent knows about your specific business, drawn from your intelligence layer, the organized body of knowledge it can reach.
Intelligence is general. Knowledge is what makes the agent’s thinking specific to you. Your processes, your context, your history, your way of doing things. Without it, the agent reasons brilliantly about nothing in particular and produces generic results. With it, the same reasoning is applied to your actual business, and the results become specific and correct.
This is why building the knowledge has to come first. The knowledge part of every agent is the intelligence layer, reached through retrieval. An agent’s knowledge is exactly as good as the intelligence layer beneath it, which is why we built that foundation before building the workers that stand on it.
Memory: What It Remembers Over Time
The third part is memory. This is what the agent retains across time, so it is not starting fresh on every single interaction.
A raw model has no memory beyond what is in front of it in a single conversation. An agent needs more than that to be a real worker. It needs to remember what it did before, what it learned, what happened last time, the ongoing context of its work. Memory is what lets an agent build on its own history instead of treating every moment as the first.
Without memory, an agent is like a worker with amnesia, competent in the moment but unable to learn or carry anything forward. With memory, it accumulates context and gets better at its job over time, the way a real employee does. Memory is what turns a series of disconnected actions into continuous work.
Skills: What It Can Actually Do
The fourth part is skills. These are the specific things the agent can actually do, the tools it can use to take real action in the world.
Intelligence lets it think. Knowledge lets it think about your business. But to do work, the agent needs the ability to act, and that is skills. The capacity to send the message, update the record, process the order, use the tool, take the step. Each skill is a specific, concrete ability to do something real, not just talk about it.
Skills are what make an agent a doer rather than a talker. An agent with great intelligence and knowledge but no skills is back to being a chatbot, able to tell you what should happen but unable to make it happen. The skills are what let it actually perform the work, and and they arguably matter even more than the intelligence.
Awareness: Knowing Its Job And Its Limits
The fifth part is awareness. This is the agent’s sense of what its job is, how it is doing, and where its limits are.
Awareness is what keeps an agent on task and out of trouble. It is knowing what it is responsible for, recognizing when it is succeeding or failing, and, critically, knowing when it has reached the edge of its competence and should escalate to a human. This is the trait we said makes an agent employable, and it is also the awareness that stands in contrast to raw intelligence. The agent supplies the intelligence. Awareness is what gives that intelligence a sense of its own situation.
Without awareness, an agent does not know its job, cannot tell when it is going wrong, and has no idea when it is out of its depth, which makes it dangerous. With awareness, it stays on task and raises its hand at its limits. Awareness is the part that makes an agent safe to trust.
Why All Five Matter Together
The reason to learn the five parts is that an agent needs all of them, and missing any one cripples it in a specific, predictable way.
Without intelligence, it cannot reason. Without knowledge, it is generic, ignorant of your business. Without memory, it cannot learn or carry context, starting fresh every time. Without skills, it can only talk, not act. Without awareness, it does not know its job or its limits, and becomes a confident danger. Each missing part produces a recognizable failure, which means when an agent disappoints you, you can usually trace it to one of the five being weak or absent.
A great agent is not one with extraordinary intelligence. It is one where all five parts are present and well-built, working together. That balance, not raw brainpower, is what makes a digital worker actually work.
What This Looks Like In Practice
Picture an agent meant to handle a recurring task, built two ways.
The first has strong intelligence and nothing else. It reasons well, but it knows nothing specific about your business, forgets everything between interactions, cannot take any real action, and has no sense of its job or limits. It is a brilliant mind in a jar, impressive and useless, able to discuss the task but not to do it, and unaware when it is wrong.
The second has all five parts. It reasons well, knows your business through the intelligence layer, remembers its ongoing work, has the skills to take real action, and is aware of its job and its limits. It actually does the task, correctly, learns as it goes, and escalates when it hits something beyond it. Same underlying model. The difference is that the second one was built with all five parts, and that is the difference between a demo and a worker.
Where To Begin
This week, take one task you wish an agent could handle and map it against the five parts.
What intelligence does the task need? What knowledge about your business would it require, and is that knowledge captured anywhere? What would it need to remember over time? What skills, what actual actions and tools, would it need to do the work? And what would awareness look like, how would it know its job and recognize when to escalate?
Working through the five parts for a real task does two things. It shows you concretely what building an agent for that task would involve, and it usually reveals which part is your weak spot, most often knowledge or skills. You are not building anything yet. You are learning to see an agent as the sum of its parts, which is exactly how you will reason about building one when the time comes.