Skip to content
Agentic Engineering

Getting Started with Agentic Tools

Pick one tool and learn it well before trying everything.

Choose your first tool

Section titled “Choose your first tool”

New to AI coding? Start with GitHub Copilot or similar. Low risk, immediate value.

Ready for more autonomy? Try a task-level agent like Cursor, Cline, or Kilo Code for multi-file changes.

Exploring? Most tools have free tiers. Try a few, commit to one for deep learning.

Your first session

Section titled “Your first session”

Start small—don’t generate your whole project.

Good first tasks:

  • Generate a single function from a clear description
  • Write tests for existing code
  • Add documentation to confusing code
  • Refactor a small, messy piece

Watch how it works: Notice what context it uses, how it handles ambiguity, where it makes mistakes.

Build prompting intuition

Section titled “Build prompting intuition”

Be explicit: Instead of “fix this bug,” try “The function calculateTotal returns NaN when items array is empty. Add a check that returns 0.”

Provide context: “This is a React component using TypeScript. Follow the pattern in other components in this folder.”

Set constraints: “Don’t modify the public API. Keep backward compatibility.”

When to delegate

Section titled “When to delegate”

Not every task should go to an agent. Ask yourself:

  1. How clear is the task? Vague tasks fail. Clear tasks succeed.
  2. How much context is needed? Deep domain knowledge is risky to delegate.
  3. What’s the blast radius? Mistakes in critical paths cost more to fix.
  4. How long would I take? If it’s 5 minutes manually, prompting might not be worth it.

Good candidates

Section titled “Good candidates”
Task typeWhy it works
Boilerplate (CRUD, DTOs, config)Repetitive, well-defined, low-risk
TestsSelf-validating—you know immediately if they work
DocumentationEasy to verify accuracy
Mechanical refactoringRenaming, extracting functions, syntax conversions
Bug fixes with clear repro”When X happens, Y occurs, but should be Z”

Keep for yourself

Section titled “Keep for yourself”
  • Architectural decisions — Agents don’t understand your system’s history or future
  • Security-sensitive code — Cost of subtle errors is too high
  • Performance-critical paths — Agents optimize for correctness, not speed
  • Novel algorithms — Agents pattern-match; new problems need human creativity
  • Ambiguous requirements — Clarify before delegating

The gray zone

Section titled “The gray zone”

For tasks that don’t fit cleanly: start with the agent, prepare to take over. Get initial structure from the agent, then refine manually.

Build habits

Section titled “Build habits”

Week 1: Boilerplate and tests only
Week 2: Add documentation and refactoring
Week 3: Feature implementation with clear specs
Week 4: Complex, multi-step tasks

Know when to stop

Section titled “Know when to stop”

Signs you should code it yourself:

  • You’ve reprompted 3+ times without progress
  • The task requires deep context the agent doesn’t have
  • You could finish manually in the time spent prompting

There’s no shame in manual coding. The goal is productivity, not agent usage.

Resources

Section titled “Resources”

Essential

Section titled “Essential”

Deep dives

Section titled “Deep dives”

Courses

Section titled “Courses”