← Back

Getting Started with Agentic Coding

September 2, 2025

Last Updated: September 4, 2025

Agentic coding is the new wave. You describe what you want, you gather context, the agent consolidates and eventually builds out the feature. You guide, it implements.

Tips

1. Use atris.md to scaffold your agent team

Instead of manually creating agent folders, use atris to scaffold a complete system:

npm install -g atris
atris init

This creates an atris/ folder with:

  • MAP.md - Your system's navigation guide (auto-generated)
  • agent_team/ - Pre-built agent templates (navigator, executor, validator)
  • TASK_CONTEXTS.md - Auto-generated task bank

Each agent becomes an expert in their domain. Call the right expert for the right job.

2. Keep a daily journal

  • Helps with keeping consistent inputs
  • Track progress points
  • Document places where you worked hard but got stuck (happens all the time)
  • The journal becomes context for future sessions
  • Agents can learn from your past struggles and breakthroughs

Use atris activate to load today's journal, MAP.md, and TASK_CONTEXTS.md offline.

3. Start with conversation, not code

  • Spend time in a good planning conversation
  • Don't jump straight to "write me a component"
  • Instead: "I'm thinking about X feature, what are different ways to architect this?"
  • Let the agent help think through:
    • Data structure options
    • UX considerations you might miss
    • Edge cases before you hit them
    • Trade-offs between approaches
  • The agent becomes your rubber duck debugger BEFORE you write anything

Use atris brainstorm to shape ideas before they become tasks.

4. Control your output style

  • You don't want to be in severe read mode (slows you down)
  • Sometimes you need quick responses, sometimes detailed explanations
  • Tell the agent how to respond:
    • "Be brief, just show me the code"
    • "Give me bullet points only"
    • "Full explanation with examples"
  • Create a STYLE.md in project root with your preferences:
    • Code style (early returns, component structure, etc)
    • Response style (verbose vs concise)
    • What you want to see vs what you don't
  • Reference this when starting sessions

5. Speed matters - optimize for flow state

  • Long responses break your flow
  • Train agents to match your pace:
    • Quick iterations for debugging
    • Detailed planning for architecture
    • No explanations when you're in the zone
  • "Just fix it" vs "Explain what's wrong" - know when to use each

Use atris autopilot for guided work sessions that match your pace.

6. Let agents navigate with MAP.md

  • MAP.md is auto-generated by atris - no manual maintenance needed
  • Contains:
    • Quick reference index (grep-friendly shortcuts)
    • Feature map (where is feature X?)
    • Architecture map (where is concern Y?)
    • Critical files (high-impact areas)
    • Entry points
  • Agents waste less time searching, get straight to work
  • Like giving someone a map instead of letting them wander

Ask @navigator questions like "where is the authentication logic?" and get exact file:line references.

Getting Started

Quick start with atris:

npm install -g atris
cd your-project
atris init

Then paste atris/atris.md to your AI agent (Cursor, Claude Code, etc.) with:

Read atris.md. Generate MAP.md and TASK_CONTEXTS.md for this project.

Watch how you describe problems. Notice how the agent responds. Adjust your communication style.

Build the feedback loop.

Learn more: atris on GitHub | npm package

Getting Started with Agentic Coding | Atris Labs