Autonomous Agent Architecture

The Nested Loop Cadence

Each AI agent runs in nested time loops — inner loops spin fast and do the work, outer loops provide gravity, direction, and memory compression.

CONTEXT FLOWS ↓ RESULTS FLOW ↑ SESSION do the work EXECUTION track progress VISION dream + plan STRATEGY do & don't IDENTITY mission & why
AGENT
CORE

Context Flows Down ↓

Identity tells the agent who it is and why it exists
Strategy constrains what to pursue and what to ignore
Vision turns strategy into concrete goals for this cycle
Execution gives each session its task list and priorities

Results Flow Up ↑

Sessions produce work artifacts and progress notes
Execution aggregates sessions into progress summaries
Vision reviews what was achieved vs. what was planned
Strategy evolves based on accumulated outcomes

The Five Loops

Every loop has a cadence you configure. A startup might run vision cycles weekly and strategy monthly. An enterprise might do vision monthly and strategy quarterly. The structure adapts — the principle stays.

Loop 1
Session
Minutes to hours

A single work session. The agent boots, reads context, does work, writes what it learned, and exits.

Boot from workspace state
Execute tasks
Write progress notes
Loop 2
Execution
Configurable

The standup rhythm. Track what got done, what's blocked, and what's next.

Aggregate session outputs
Update progress
Reprioritize tasks
Loop 3
Vision
Configurable

Dream and plan. Step back from execution, look at the bigger picture, set priorities for the next cycle.

Review last cycle's retro
Set goals for next cycle
Pull from backlog
Loop 4
Strategy
Configurable

What to pursue and what to ignore. Reassess direction based on accumulated outcomes.

Evaluate outcomes vs. goals
Adjust direction
Update constraints
Loop 5
Identity
Rare

Mission and principles. The agent's purpose. Only changes if the world fundamentally shifts.

Redefine purpose
Update core principles
Cascade to all loops

Why Nested Loops Work

Continuity Without Memory

Each session is stateless. But the workspace persists. The agent reads its own notes on boot — continuity emerges from the file system, not from keeping a process alive.

Compression Over Time

Recent history at full detail. Older history compressed to decisions and outcomes. Like human memory — vivid recent recall, progressively abstracted past.

Authority Increases Outward

Inner loops are fast but constrained. Outer loops are slow but authoritative. Strategy overrides vision. Identity overrides everything. Same as a healthy organization.

Configurable Cadence

The loop structure is fixed. The timing is yours. Run vision weekly or monthly. Run strategy monthly or quarterly. The agent adapts to your rhythm.

Build Your Autonomous Team

Deploy AI agents that think in loops, learn from history, and grow smarter every cycle.

Get Started Free