PTENES
← Track 6 | Module 6.1
Module 6.1

Planning Your Assistant

Scope definition, architecture selection, identity design, and build roadmap.

1

What Is AI Assistant Planning?

Planning an AI assistant is different from planning traditional software. You are defining not only functionality, but behavior, identity, and ethical boundaries of an agent that will act on your behalf.

The 4 Planning Dimensions
1Functional scope: what the assistant CAN and CANNOT do
2Identity: who the assistant is, its values, and its communication style
3Architecture: which Claw pattern and which components are needed
4Roadmap: sequence of steps to build from the MVP to the complete assistant
2

Why Plan Before Coding?

In AI systems, the cost of late changes is much higher than in traditional software. Changing the architecture after memory, tools, and channels have been configured may require complex migrations.

Decisions that are costly to change later
  • · Memory database schema
  • · SOUL.md format
  • · Base Tools interface
  • · Channel authentication standard
Decisions that are easy to change
  • · AI model used
  • · SOUL.md communication style
  • · Adding new Tools
  • · Channel settings
3

Defining Scope and Use Cases

Use the Jobs-to-be-Done technique to map what you want to delegate to the assistant. Be specific: "manage my email" is vague; "filter and categorize work emails, automatically responding to scheduling requests" is a concrete job.

Job Definition Template
WHEN I receive a customer email requesting a meeting
I WANT YOU TO the assistant checks my calendar and suggests times
WHY I don’t have to do this triage manually
CONSTRAINTS: never accept without my final approval
CHANNEL: Gmail + Google Calendar
ARCHITECTURE: IronClaw (sensitive data)
4

Choosing the Right Claw Architecture

Use this decision tree to choose the right pattern based on your security, scale, and complexity requirements.

Sensitive third-party data?
  YES → IronClaw
  NO → continue...
Multiple specialized agents?
  YES → TinyClaw
  NO → continue...
Should the system improve autonomously?
  YES → Agent Zero
  NO → continue...
Offline or edge environment?
  YES → ZeroClaw or NanoClaw
  NO → OpenClaw (use it and refine it later)
5

Identity Design and SOUL.md

SOUL.md is the most important document for your assistant. Give it serious time — a well-defined identity prevents inconsistent behavior and out-of-context responses.

# SOUL.md template for final project
# Identity
Name: [YOUR ASSISTANT]
Role: Personal assistant specialized in [YOUR CONTEXT]
# Personality
Tone: [direct/formal/casual/technical]
Strengths: [list the top 3-5]
# Values and Constraints
Always: [required behaviors]
Never: [prohibited behaviors]
# Context
User: [who you are and your professional context]
6

4-Phase Development Roadmap

Don't try to build everything at once. Follow the 4 evolutionary phases to get an assistant working quickly and improve it with real feedback.

1
MVP Phase (week 1-2)
One channel, one model, basic SOUL.md, no external tools. Focus: answering questions and maintaining context.
2
Core Phase (week 3-4)
Persistent memory, 2-3 essential tools, identity refinement based on real-world use.
3
Advanced Phase (month 2)
Multiple channels, automations, integrations with Google/GitHub, security hardening.
4
Autonomous Phase (month 3+)
Multi-agents, Agent Zero loop, automatic tool discovery, complete monitoring.
← Back to Track 6 Course Conclusion