Agentic Chicago • Learn • Series Part 1 of 3
Building a General AI Agent, Part 1: Context Initialization and Domain-Selective Loading
If you want one agent to handle both sales and engineering work, start with context discipline. Keep one stable identity. Keep startup context lean. Load domain expertise only when the task requires it.
What a general agent really is
A general agent is not a giant prompt with every playbook loaded all the time. It is one consistent operating identity that can switch domains without dragging irrelevant context into the current task.
The core design principle
Separate context into three layers:
- Core startup context loaded every session.
- Domain context loaded only after task classification.
- Task context loaded for the active request only.
How domain loading works in practice
- Agent receives request.
- Router classifies domain and intent.
- Runtime loads only the required domain pack.
- Router references specialist skills for execution details.
- Agent executes with validation gates.
Bare-bones checklist
- One stable agent identity.
- One startup context pack with non-negotiable behavior rules.
- One domain router layer.
- Specialist skills loaded by router only.
- Explicit precedence rules.
Need this implemented in your workflow?
If you want this architecture set up for your team, book a strategy session.