Training Outline: Working With AI Agents
This guide teaches a simple shift: stop treating AI as a magic answer machine and start using it as a system for repeated work.
The core idea is straightforward. Much of your workday is not pure judgment. It is gathering, filtering, reformatting, summarizing, routing, and preparing information. That work matters, but it is mechanical. It follows patterns. Once you can see those patterns clearly, you can design workflows where an agent handles the repeated mechanics and you keep control of the parts that require judgment, taste, relationships, and accountability.
The central thesis
AI is most useful when it handles repeatable steps so you can focus on higher-value human work.
This is not a guide to replacing yourself. It is not a guide to automating everything. It is a guide to building better systems around your own work.
Goal of this session
Learn how to break work into repeatable steps, then decide which steps should stay human and which steps an agent can help execute.
Part 1 — What AI actually is
How Large Language Models Work
Large Language Models (LLMs) are trained on massive amounts of text data (books, websites, documentation, conversations).
They work by predicting the next piece of text based on patterns they learned.
In simple terms:
LLMs are extremely powerful pattern recognition and pattern completion machines.
Because of this, they are good at tasks that follow recognizable patterns.
Tasks that follow patterns can often be automated
Examples:
- Gathering information
- Filtering data
- Sorting information
- Transforming formats
- Summarizing content
- Monitoring events
- Categorizing inputs
- Generating drafts
- Extracting structured data
These tasks show up in almost every job.
Part 2 — What humans still do best
AI is powerful, but it should be seen as a tool for execution, not a replacement for human judgment.
Humans should focus on:
- Creativity
- Relationships
- Judgment
- Strategy
- Taste
- Trust
- Empathy
- Responsibility
- Decision making
A helpful way to think about this:
Humans design the system.
AI helps run the system.
Part 3 — Seeing your work as workflows
Most jobs feel messy and unpredictable, but when you look closely, they are made of repeatable workflows.
Example:
Responding to a customer request
- Receive message
- Identify request type
- Gather relevant information
- Draft response
- Format response
- Send response
This is a workflow.
Why this matters
AI agents operate best when they are given:
- Clear steps
- Defined inputs
- Defined outputs
- Repeatable processes
So the key skill is learning to see your work as a sequence of steps.
Part 4 — How to deconstruct a workflow
To automate work, we break it down into simple operations.
Questions to ask:
- What triggers this work?
- What information is needed?
- What steps happen every time?
- What decisions are made?
- What is the final output?
Example Deconstruction
Market research task
Original task: "Research competitors and summarize findings"
Broken into steps:
- Gather competitor websites
- Extract product descriptions
- Identify key features
- Compare positioning
- Summarize insights
- Format report
Each of these steps can potentially be handled by AI agents.
Part 5 — The core components of AI agents
When we build an AI agent system, there are a few key pieces.
1. LLM (The Brain)
The LLM handles:
- reasoning
- writing
- classification
- summarization
- decision making
But the LLM does not inherently know your business or access your systems.
That's where the other components come in.
2. Tools (The Hands)
Tools allow agents to interact with systems.
Examples:
- Search the web
- Access databases
- Read documents
- Send emails
- Update spreadsheets
- Call APIs
- Monitor systems
Tools allow the AI to take actions, not just produce text.
3. Context (Short-Term / Working Memory)
Context is the information the agent can see while working.
You can think of it as the agent's short-term or working memory for the current task.
This can include:
- instructions
- company knowledge
- documents
- previous messages
- data from tools
Relevant context leads to better outputs.
For the practical side of managing context during real work, review Working with agents.
4. Loops (The Work Cycle)
Agents often operate in loops:
- Observe information
- Think about next step
- Use a tool or generate output
- Evaluate results
- Repeat until task is complete
This allows agents to solve multi-step problems.
Part 6 — Designing agent workflows
The real power comes from combining these components into automated systems.
Basic structure:
Trigger → Agent → Tools → Output
Example:
Customer Email Arrives
↓
Agent classifies request
↓
Agent gathers data from CRM
↓
Agent drafts response
↓
Human reviews (optional)
↓
Email sent
Part 7 — Practical exercise
We will now apply this to your real workflows.
Steps:
- Identify a repetitive task in your role
- Break it into steps
- Identify which steps follow patterns
- Determine which steps could be automated
- Design a simple agent workflow
Key Takeaways
- AI excels at pattern-based work
- Most jobs contain repeatable workflows
- Automation starts by breaking work into steps
- AI agents combine:
- LLMs (thinking)
- Tools (actions)
- Context (knowledge)
- Loops (execution)
Your role becomes:
Designing the system that AI helps run.