Agentic Coding

Day 1 Exercises: The Landscape

Setup

  • Each attendee needs a laptop with terminal access
  • Claude Code installed (or ready to install -- Exercise 1 covers installation)
  • A small existing project to work with (personal project, open source clone, or provided sample). The project should have at least 5-10 source files and a test suite.
  • Access to at least one other AI coding tool for comparison (Copilot, Cursor, or similar). If unavailable, the facilitator demonstrates.

Exercise 1: Installation and First Contact (20 minutes)

Objective: Get Claude Code running and complete a first agentic interaction.

Instructions:

  1. Install Claude Code following the official setup guide.
  2. Open a terminal in your project directory.
  3. Ask Claude Code to explain the structure of your project. Note what it gets right and what it misses.
  4. Ask it to add a small feature or fix a minor issue -- something you already know the answer to.
  5. Review the result. Did it do what you expected? What surprised you?

Facilitator notes: The first interaction is about calibration, not productivity. Attendees should compare what they expected with what happened. Common surprises: the agent reads more context than expected, it may propose changes to files you did not mention, it may ask clarifying questions. All of these are features of agentic behaviour, not bugs.

Verification: The attendee can describe one thing the agent did that they did not explicitly ask for.

Exercise 2: The Spectrum Self-Assessment (15 minutes)

Objective: Place yourself on the agentic-assistive spectrum and identify your current ceiling.

Instructions:

  1. Review the spectrum table from ACI-015:
    • Level 1: Autocomplete (accept suggestions as you type)
    • Level 2: Chat/Q&A (ask questions, get explanations)
    • Level 3: Multi-file editing (agent modifies multiple files in one task)
    • Level 4: Autonomous multi-step (agent plans and executes across sessions)
  2. For each level, write down one task you have done at that level (or "never" if you have not).
  3. Identify your current ceiling -- the highest level you use regularly.
  4. Identify one task from this week's work that could have been done at a higher level than you used.

Facilitator notes: Most attendees will place themselves at Level 1 or 2. A few will be at Level 3. Level 4 is rare outside agentic-native practitioners. The point is not judgement -- it is awareness. The "one task that could have been higher" question seeds Day 2's practical work.

Verification: The attendee can state their current level and name one task they could try at a higher level.

Exercise 3: Assistive vs Agentic Comparison (30 minutes)

Objective: Experience the difference between assistive and agentic tool use on the same task.

Instructions:

  1. Choose a task in your project: add a new function, refactor an existing one, or write tests for an untested module.
  2. First, attempt the task using an assistive tool (Copilot autocomplete, or Claude Code in chat-only mode). Accept suggestions, ask questions, but do the integration yourself.
  3. Note how long it took and how much you did manually.
  4. Now attempt a similar task (different function, different test) using Claude Code in agentic mode. Describe the outcome you want and let the agent plan and execute.
  5. Compare: What did the agent do differently? Where was it better? Where did you have to intervene?

Facilitator notes: The comparison should produce a visceral "aha" moment for attendees who have only used autocomplete. The agentic attempt will not be perfect -- that is part of the lesson. The agent will do things the attendee did not expect (read files, propose structural changes, create new files). The discussion should focus on what changed about the attendee's role: from typist to reviewer.

Verification: The attendee can articulate one concrete difference between the assistive and agentic experience.

Stretch Goals

  • Review the evaluation framework preview and score your current tool setup across the 6 dimensions. Bring your scores to Day 5.
  • Read one ACI from Day 2 (ACI-017: The Delegation Model) as preparation for tomorrow.
  • If you have access to multiple AI coding tools, repeat Exercise 3 with a different tool and compare the agentic capabilities.