⚡ What a skill is
A skill is a specific, actionable capability of an agent—something concrete it knows how to do and can be triggered on demand. “Summarize a text,” “classify an email,” and “look up a spreadsheet” are skills. Notice the precision: it’s not “be intelligent” or “help the customer” (those are roles or goals), but a bounded action with clear inputs and outputs. Thinking about skills means thinking about verbs, not in adjectives.
🎯 A skill is a verb, not an adjective
"Attentive" and "competent" aren't skills—they're impressions. A skill has to fit into a sentence like "it can do X with Y and return Z". If you can't describe the input and output, it isn't a skill yet: it's a vague intention waiting to be made concrete.
✓ This is a skill
- ✓Summarize a contract in 5 lines
- ✓Classify a ticket by urgency
- ✓Check a product's inventory
- ✓Generate a proposal from a brief
✗ This isn't a skill
- ✗"Be helpful"
- ✗"Understand the customer"
- ✗"Be intelligent"
- ✗"Resolve everything" (open-ended scope)
🧰 Common skills
Most AI solutions are built by combining four skill families: summarize, classify, look up, and generate. Summarizing condenses a lot of information into a little. Classifying puts each thing into a category (urgent/normal, hot/cold lead). Looking something up finds a specific piece of data where it lives (a spreadsheet, a system, a database). Generating creates something new from what it received (a text, a proposal, a response). Almost every business problem is a recipe made from these four.
Summarize
A lot of information goes in; the essentials come out. Contracts, meetings, long emails.
Classify
Each item gets a label. Urgency, topic, sentiment, request type.
Consult
Find a specific piece of data where it lives. Inventory, order status, price.
Generate
Produce something new from the context. Proposal, response, draft.
The 4 families, one line each
Illustrative list—every skill has an input and an output.
🎯 Skill tied to an objective
Every skill only makes sense when it’s tied to a clear goal. The question isn’t “what cool capability can I add?” but “what goal is this agent pursuing, and which capability brings it closer?” The capability is the means; the goal is the end. When you start with the goal, every capability has a reason to exist—and it’s easy to tell when one is unnecessary.
💡 Practical tip
Before adding any skill, write the sentence: "this skill exists so the agent can ___ (goal)." If you can’t complete the sentence with a real business goal, the skill shouldn’t be included—not yet.
🚫 Without intent, it becomes excess
There’s a strong temptation: to give the agent all all possible skills, "just in case." It's a mistake. A skill without intent becomes bloat, and excess comes at a cost: the agent gets slower, more expensive, harder to predict, and more likely to use the wrong skill at the wrong time. More skills do not mean more value — they often mean more confusion. A useful agent is lean: it has exactly what it needs to accomplish its goal.
✓ Lean agent
- ✓Only the skills that serve the goal
- ✓Fast and predictable
- ✓Easy to explain what it does
- ✓Choose the right skill without hesitation
✗ Bloated agent
- ✗Skills "just in case," with no real use
- ✗Needlessly slow and expensive
- ✗No one really knows what it does
- ✗Uses the wrong skill at the wrong time
💡 Practical tip
The pruning question: "if I remove this skill, will the agent stop meeting any real objective?" If the answer is no, remove it. An agent with 4 well-chosen skills is more valuable than one with 12 generic skills—less is more when every piece has a purpose.
🎚️ Activate the Right Skill
Having the skill isn’t enough—the agent needs to trigger the right skill at the right time. When someone sends a contract and asks “what’s important here?”, the answer is to summarize, not generate. When they ask “is it in stock?”, you look it up, you don’t classify it. This routing—reading the request’s intent and choosing the right skill—is what separates an agent that gets it right from one that answers with anything. The request’s intent guides the choice.
The request arrives
The user speaks in natural language—not always specifying which skill they want.
Reads the intent
The agent interprets what the person really wants to achieve with that request.
Choose the skill
Summarize? Classify? Look up? Generate? Choose the one that fits that intent.
Executes and returns
Triggers the skill, produces the output, and responds—actually resolving the request.
🧭 The same topic, different skills
- "What matters in this contract?" → summarize.
- "Is this contract urgent, or can it wait?" → classify.
- "What is the total value of this contract?" → look it up.
- "Write an email responding to this contract." → generate.
🧩 Skills make up a useful agent
In the end, a useful agent is a lean composition of skills tied to a goal. It’s not the number of skills that creates value—it’s how well they align with the agent’s purpose. A good customer service agent, for example, usually combines a few well-chosen skills: classify the request, look up the information, and generate the response. Each one brings the agent closer to its goal. Skills are the pieces; intent is what puts them together in the right order.
An agent's customer support skills
Illustrative outline — 4 skills, all serving the same goal.
🧠 Intent assembles the pieces
The previous modules gave the system its soul (who it is) and its agents (who does the work). Skills are what each agent knows how to do. But what determines which skills each agent has, and in what order it uses them, is still the intent at the center — the same thread that weaves through the entire course.
Self-check (optional): why is giving an agent every possible skill a mistake?
⚡ Module summary
Next module:
2.5 — Memory and Context: what Jarvis remembers so he can act with history, not in the dark.