You do not need engineers, a big budget, or a single line of code to put an AI agent to work. Our latest webinar proved it, going from a blank screen to a working tool in one sitting. Sean Sanker Jr., our Director of AI and Business Development, broke down what an agent is and showed how to build an AI agent with Claude in a few minutes of conversation.
If you missed the session, you can watch the full recording below.
What really is an agent?
Sean started by grounding the term “agent” in something everyone already uses. A chatbot answers a question. An agent takes a goal and keeps working until it is done. Under the hood it runs on three parts. The model is the brain that reads, reasons, and decides the next step. The tools are its hands, the files, calendar, and systems it can actually touch. The instructions are the playbook for how your team wants the work done. The model and the tools come with Claude. The instructions are the part you supply.
If you’ve used AI lately, you have seen this in action. Ask for the latest news and it does not answer from memory. It decides it needs more information, searches the web, reads what it finds, and only then writes you a summary. Ask for the five most recent papers on a topic and it will fetch each one, summarize it, and keep going until the task is complete. That loop, model plus tools plus instructions, is the whole idea. If you use Claude today, you are already using an agent.
So when people say they are “building an agent,” they are usually not building that loop but customizing it. The people building agents from scratch work at companies like Anthropic and OpenAI. The rest of us shape how an existing agent behaves, and we do that with skills.
What is a skill?
Simple, skills are just text. A skill is a plain text file with a name, a short description of what it does, and a set of instructions.
Think of a skill as your process, written down once and packaged so Claude can pull it in whenever it is relevant. Instead of retyping the same detailed prompt every time and getting slightly different results, you write it down once and get far more reliable output. It does not make AI perfectly predictable, but it removes most of the guesswork.
A few things made skills click for the audience:
- They are reusable building blocks. You can keep a small library for things like brainstorming, project planning, and reviewing work, and pulls in whichever one fits the moment.
- Claude can load them automatically. The name and description always sit in the background, so the agent knows what it has available and can reach for the right one mid-conversation. The description is the part Claude reads when it decides whether to pull your skill off the shelf, so it is worth writing clearly.
- They are not locked to one tool. Skills started with Anthropic and became an open standard, so the same idea now works in Claude and, on business tiers, in ChatGPT and Microsoft Copilot.
What makes a good candidate for a skill
You can turn almost any repeatable process into a skill. Sean offered a simple self-audit. Ask yourself these questions about your week:
- What do you do on a recurring basis that follows roughly the same steps?
- Where are you moving information by hand from one place to another, like copying data between spreadsheets or files between SharePoint sites?
- What tedious task do you keep putting off or rushing at the last minute?
Good starting points include weekly status reports, turning a meeting into action items, and running a formatting or quality check on a document before it goes out. Skills do not have to be rigid either. Open-ended work like brainstorming is a great fit, and newer models often perform better when you give them room rather than scripting every step.
How to build a skill
This was the most practical part of the session, and it is where our AI deployment work with clients usually starts. Sean’s method is low-effort:
- Brain dump first. Use voice-to-text and just talk through how you do the task. Most people speak faster than they type and explain more when they ramble than when they write.
- Let Claude interview you. Paste in the messy transcript and ask Claude to question you, spot gaps, and prototype a few layouts before anything gets written.
- Test on real work. Run a real input through the skill, see where it falls short, then hand the result back to Claude and explain what went wrong so it can refine the instructions.
- Share it with your team. A good skill helps everyone who does the same task, and Claude lets you share skills across a team.
To make that first brain dump easy, Sean shared a fill-in-the-blanks prompt. Copy it, fill in the bracketed parts, and send it to Claude to build your first skill:
Help me create a skill for [the task].
Here is how I do it today: [walk through your steps and decisions].
I start with [your inputs, like meeting notes or a data export], and the final output should be [what you produce, and in what format].
Follow these standards: [tone, formatting rules, things to avoid].
Here is an example of a great final result: [paste or attach one].
Ask me clarifying questions before writing the skill.
Claude will ask a few questions, draft the skill, and you refine from there.
Sean built a meeting debrief skill live on the call. It turns a raw meeting transcript into a scannable ledger: what was decided, what was left open, what got pushback, and a short list of action items with owners. The whole build took a few minutes of talking.
One word of caution
Because a skill can direct an agent to take real actions, treat it with the same care you would any tool. Sean’s advice was direct: do not download and run random skills you find online without reading them first. Just as important, have the AI check its own work. Asking a model to verify its own results is one of the most underrated habits in getting reliable output, and it matters even more once agents start doing multi-step work on their own.
Want help building your first skill?
Sean covered far more in the session, and you can watch the full recording to learn more about skills and how to build them. For more on putting agents to work responsibly, check out our recap of the Strategic AI webinar.
If you would like to build an AI agent with Claude for a real process on your team, book a free working session with us. We will help you pick the right first skill and get it working. If you’d like to join us for our next webinar, sign up for our newsletter and we will let you know when the invite goes live.
