Agentic Tools Training

Get up and running with Claude Code in an afternoon.

A self-guided track designed to help people without an engineering background move beyond the chat interface and start using AI agents and agentic workflows to research, automate, and build.

Free. No sign-up required. Takes just a few hours to complete.

0% 0 of 30 set up
The plan

Three short phases: set up your tools, read through some terms and concepts that you will encounter again and again when working with AI agents, then put an agent to work on a real task.

Phase 0

Before your first session

Create the accounts and install the tools you'll use most. Each takes a couple of minutes. Tap a row to check it off.

Phase 1

Concepts worth knowing

A quick glossary — read each one, then mark it understood and it'll tuck itself away. These are the concepts you'll encounter again and again when using AI agents to research, automate, and build.

.gitignore & secrets

Git defines a .gitignore file as one that “specifies intentionally untracked files that Git should ignore” — like build output, dependencies, and secrets. Pair it with a .env file for keys, and never commit that. Git ↗

.env files

A .env file is a plain-text file that stores a project’s configuration values and API keys. The agent reads keys from it, while .gitignore keeps the file out of version control so secrets stay private.

The terminal / CLI

The terminal (command-line interface, or CLI) is a text-based way to run programs by typing commands instead of clicking. Agentic coding tools run here.

Markdown & session-progress.md

Markdown’s creator John Gruber describes it as a “text-to-HTML conversion tool” with an “easy-to-read, easy-to-write plain text format.” A session-progress.md file uses it to let a long task record progress so work can resume. Daring Fireball ↗

Plan mode

Anthropic describes plan mode as a permission mode where Claude Code “reads files and runs read-only shell commands to explore but does not edit source files” — you review its proposed plan, then approve execution. Anthropic ↗

Agents & parallelism

Anthropic describes subagents as “specialized AI assistants that handle specific types of tasks,” each running in its own context window. You can spin up several to work on a problem and combine their results. Anthropic ↗

MCP

The Model Context Protocol docs describe MCP as “an open-source standard for connecting AI applications to external systems” — “like a USB-C port for AI applications” for plugging in tools and data sources. Model Context Protocol ↗

Skills & plugins

Anthropic describes Skills as a way to “extend what Claude can do”: you create a SKILL.md file with instructions and Claude adds it to its toolkit, using it automatically when relevant. Anthropic ↗

Hooks & guardrails

Anthropic describes hooks as “user-defined shell commands … that execute automatically at specific points in Claude Code’s lifecycle” — letting you add deterministic checks and guardrails around the agent’s actions. Anthropic ↗

Permissions & auto-accept

Claude Code “pauses and asks you to approve” actions like editing files or running commands; permission modes such as acceptEdits let you auto-approve routine actions — but never blanket-approve destructive ones. Anthropic ↗

Hallucinations are real

OpenAI defines hallucinations as “plausible but false statements generated by language models.” Because the model can be confidently wrong, keep a human in the loop and build fallbacks. OpenAI ↗

Interrupting (the Escape key)

Pressing the Escape key stops the agent mid-action so you can correct course before it goes further down the wrong path.

Structured vs unstructured data

Structured data has a clear, predictable shape (tables, JSON, spreadsheets), while unstructured data is free-form (prose, emails, screenshots). Agents handle structured data more reliably.

Deterministic vs not

A deterministic step returns the exact same result every time it runs with the same input, while a language model’s output is non-deterministic and can vary between runs.

Usage limits & token budget

OpenAI explains that models process text as “tokens” and that “API usage is priced per token.” Agentic runs consume tokens that cost money, so reach for cheaper tools when full autonomous execution isn’t needed. OpenAI ↗

Point it at the docs

Give the agent a link to the official documentation for a library or API. It works far better from the real, current reference than from its training memory.

Connecting Jira / Linear

Wire the agent into your issue tracker, such as Jira or Linear, so it can read existing tickets and create new ones as part of its work.

Design agents

A design-focused agent reviews a page from a URL or screenshot and gives you visual feedback on layout, spacing, and styling.

Make long jobs resumable

Have long-running work checkpoint its progress as it goes, so a crash or interruption doesn’t throw away everything done so far.

An always-on runner

An always-on machine that never sleeps gives agents a stable place to run around the clock, independent of your laptop being open.

Web search & browser tools

Anthropic describes the web search tool as giving Claude “direct access to real-time web content” beyond its knowledge cutoff. Agents can also drive a real browser to read pages, fill forms, and capture screenshots. Anthropic ↗

Version control & GitHub

Git’s site defines it as “a free and open source distributed version control system” that tracks changes to your code. GitHub stores those repositories remotely and makes them easy to share. Git ↗

Phase 2

Your first real agent workflow

Now it's time to put an agent to work. Copy each prompt into Claude Code and watch it research, browse the web, use tools, and complete real tasks on your behalf.

Research + browser + buildParallelism, web search, and browser automation in one go.
Spin up agents and use web search plus browser tools (e.g. Playwright) to
find articles published about https://www.nantucketmovingcompany.com in the
last two years. Confirm each result is about the Nantucket company specifically,
not other movers with a similar name. Write the findings to a clean, styled
.html file I can open in my browser.
Use a secret + a connected toolReading from .env and creating a real ticket via an API.
Using the JIRA_API_KEY from my .env file, create a ticket in this project
titled "Save my local work to a remote repository." Then share the ticket URL
here once it's created.
Ship itGet your work off your laptop and into version control.
Initialize a git repository for this project, create a new GitHub repo under
my account, commit everything, and push it. Walk me through what each step does
as you do it, and give me the repo URL at the end.

By the end, you'll understand how to:

  • Delegate research to multiple agents working in parallel.
  • Connect AI to tools, APIs, and your own data.
  • Hand off real tasks and review the results.
  • Use the workflows that power modern AI-assisted work.

Save your progress across devices

Your progress is already saved automatically in this browser (locally) — no account, no sign-in. Add your email and we'll send a link to pick up where you left off on any device.

Want a hand getting started?

I'll walk you through this live, on your machine, using your own projects and workflows.

Book an onboarding session →
← Back to home