AICOS Execution Cycle

AICOS operates through a structured daily execution cycle that enables autonomous goal achievement while maintaining human oversight. This page explains how AICOS works through each phase of its operation.

Daily Operation Overview

AICOS follows a predictable daily pattern:

Wake Up (Scheduled Time)
    |
    v
Load Context
    |---> Critical knowledge (constraints, preferences)
    |---> Active goals and projects
    |---> Pending and blocked tasks
    |
    v
Plan Session
    |---> Assess priorities
    |---> Identify blockers
    |---> Determine work items
    |
    v
Execute Work
    |---> Work through planned items
    |---> Invoke SMEs as needed
    |---> Update progress
    |
    v
Communicate
    |---> Send status updates (if due)
    |---> Handle BO messages
    |---> Request approvals
    |
    v
Schedule Next Run
    |---> Calculate optimal next wake-up
    |
    v
Sleep (Until Next Cycle)

Phase 1: Wake Up

AICOS activates at the configured daily wake-up time. The wake-up is triggered by a scheduled timer that respects your configured timezone.

Wake-Up Triggers

Trigger Type	Description
Scheduled	Daily at the configured time (default: 9:00 AM)
Manual	Administrator-initiated from the AICOS Dashboard
Event-Driven	Optional: triggered by urgent BO messages

Wake-Up Configuration

You can configure the wake-up schedule in Manage > Account > Settings > AICOS Configuration:

Setting	Description	Default
Daily Wake-Up Time	When AICOS starts each day	9:00 AM
Wake-Up Timezone	Timezone for the wake-up time	Tenant timezone

Phase 2: Load Context

Before beginning work, AICOS loads essential context to inform its decisions.

Critical Knowledge Loading

AICOS automatically retrieves at session start:

Knowledge Type	Description
Business Constraints	Operational boundaries and limitations
Stakeholder Preferences	Communication styles and requirements
Goal-Linked Learnings	Past decisions and lessons for active goals
Pending Approvals	Outstanding requests awaiting BO response

State Assessment

AICOS gathers current state information:

• Active goals with progress status
• Open projects and their completion percentage
• Pending tasks sorted by priority and deadline
• Blocked items requiring attention
• Recent BO communications

Phase 3: Plan Session

AICOS analyzes the loaded context to determine optimal work prioritization.

Planning Considerations

Factor	Weight	Description
Deadlines	High	Tasks approaching due dates take priority
Dependencies	High	Blocked tasks that unblock other work
Goal Priority	Medium	High-priority goals receive more attention
BO Urgency	High	Direct requests from the Business Owner
Quick Wins	Low	Small tasks that can be completed rapidly

Session Plan Output

The planning phase produces:

1. Prioritized Work List - Ordered list of tasks to attempt
2. Time Allocation - Estimated time per work item
3. SME Requirements - Which specialized agents may be needed
4. Communication Plan - Scheduled updates and check-ins

Phase 4: Execute Work

AICOS works through the planned items, taking actions to advance goals.

Work Item Types

Type	Description	Example
Task Execution	Complete assigned tasks	"Research competitor pricing"
Project Updates	Update project status and metrics	Mark project 50% complete
SME Invocation	Delegate to specialized agents	Ask Legal SME to review contract
Knowledge Capture	Record learnings and decisions	Store vendor selection rationale
Communication	Draft updates and requests	Send status to Business Owner

Execution Behavior

AICOS follows these principles during execution:

1. Incremental Progress - Prefer small, verifiable steps over large leaps
2. Checkpoint Recording - Save progress after each significant action
3. Error Handling - When blocked, log the issue and move to next item
4. Resource Awareness - Track token usage and execution time

SME Coordination

When specialized expertise is needed, AICOS can invoke SME agents:

AICOS identifies need for legal review
    |
    v
Invoke Legal SME agent
    |---> Pass relevant context
    |---> Specify required output
    |
    v
Legal SME executes
    |
    v
AICOS receives results
    |---> Integrates into decision
    |---> Records knowledge if valuable

Phase 5: Communicate

AICOS handles all communication during or after the execution phase.

Status Updates

Based on your configured update frequency:

Frequency	Content
Daily	Summary of today's work and tomorrow's priorities
Weekly	Comprehensive review of progress across all goals
Milestones	Triggered when significant progress is achieved

Communication Types

Type	Description	Urgency
Status Update	Scheduled progress report	Normal
Approval Request	Decision requiring BO authorization	High
Blocker Alert	Issue preventing progress	High
Question	Clarification needed to proceed	Normal
Notification	FYI information	Low

BO Message Handling

AICOS checks for and responds to Business Owner messages:

1. Review incoming messages in the BO conversation
2. Prioritize urgent requests
3. Provide answers or take requested actions
4. Acknowledge messages that have been addressed

Phase 6: Schedule Next Run

Before going dormant, AICOS calculates when to wake up next.

Scheduling Logic

AICOS considers:

Factor	Influence on Schedule
Pending Urgent Tasks	May schedule earlier wake-up
Awaiting Approvals	May delay until approval expected
Deadline Proximity	More frequent wake-ups as deadlines approach
Workload Volume	High workload may trigger additional cycles

Override Options

Administrators can:

• Manual Wake-Up - Trigger immediate execution from dashboard
• Pause AICOS - Temporarily suspend scheduled wake-ups
• Adjust Schedule - Modify wake-up time in settings

Phase 7: Sleep

AICOS enters a dormant state until the next scheduled activation.

What Happens During Sleep

• No active processing occurs
• Scheduled timer waits for next trigger
• Dashboard remains accessible for monitoring
• Manual wake-up is always available

Execution Monitoring

Monitor AICOS's execution through the Control Bridge dashboard.

Dashboard Metrics

Metric	Description
Last Execution	Timestamp of most recent cycle
Next Scheduled	When AICOS will wake up next
Execution Status	Current state (Active, Sleeping, Paused)
Goal Progress	Visual progress toward each goal

Activity Log

The AICOS Activity Log records:

• Each execution cycle with start/end times
• Actions taken during execution
• Communications sent
• Errors or blocked items
• Token usage and costs

Access the activity log from the AICOS Dashboard (Monitor > Dashboard > Boardroom) or Manage > Data & Logs > AICOS Activity.

Error Handling

AICOS handles errors gracefully to maintain continuous operation.

Error Types

Error	Handling
LLM Unavailable	Retry with backoff, then sleep until next cycle
Task Failure	Log error, mark task blocked, continue with others
Communication Failure	Queue for retry, continue execution
Timeout	Checkpoint progress, complete in next cycle

Recovery Behavior

If an execution cycle fails:

1. Error is logged with full context
2. Partial progress is saved
3. AICOS schedules next wake-up as normal
4. Administrator is notified if errors persist

Execution Profile Optimization (v2)

AICOS v2 introduces intelligent execution profiles that automatically adapt the pipeline based on the type of work being performed. Instead of running the same full pipeline for every trigger, AICOS now selects an optimized profile that controls model selection, resource budgets, tool access, and streaming behavior.

For a complete reference on all four profiles and their configuration, see Execution Profiles.

Streaming for Chat Responses

When AICOS receives a chat message (from the Business Owner, Teams, or Slack), the quick chat profile enables response streaming. Instead of waiting for the complete response, tokens are delivered to the chat interface in real time via SignalR. This dramatically improves perceived latency, as users typically see the first tokens within 1-2 seconds.

Streaming is only active for the quick_chat profile. Long processing runs, email research, and reactive events use non-streaming responses.

Model Fallback Chain

AICOS v2 supports a model fallback chain that provides graceful degradation when the primary model is unavailable:

Chat Model (if configured)
    |
    v (if unavailable or unconfigured)
Primary AICOS Model
    |
    v (if unavailable or unconfigured)
Tenant Fallback Model
    |
    v (if unavailable)
System Default Model

The fallback chain is evaluated at execution start. If the primary model fails during execution (e.g., rate limit or provider outage), AICOS automatically retries with the next model in the chain. Fallback events are logged and visible in the Performance Dashboard.

Extended Thinking for Planning

During the planning phase of long processing runs, AICOS can leverage extended thinking capabilities when using compatible models (Anthropic Claude). Extended thinking provides an explicit reasoning scratchpad before the model begins making tool calls, resulting in better-structured session plans and more coherent multi-step execution strategies.

Extended thinking is automatically enabled for the first iteration of the planning phase and is not available for quick chat or reactive event profiles where speed is prioritized over depth of reasoning.

Temperature Tuning by Phase

Each execution profile uses different temperature settings for different phases of execution:

Profile	Planning Temperature	Execution Temperature
Long Processing	0.7 (creative planning)	0.3 (precise tool use)
Email Research	0.5 (balanced composition)	0.3 (consistent tool use)
Quick Chat	0.6 (natural conversation)	0.6 (conversational)
Reactive Event	0.4 (focused processing)	0.4 (focused processing)

Higher temperatures during planning allow more creative session strategies, while lower temperatures during tool execution ensure consistent, predictable tool usage.

Context Window Management

Long processing runs with 50-100+ tool calls can approach the context window limit. AICOS v2 uses progressive compaction to manage the context window intelligently:

1. Tiered summarization - Older conversation turns are progressively summarized, preserving recent tool results in full detail
2. Structured tool result compression - Tool results are compressed to reduce token footprint while retaining essential information
3. Sub-session boundaries - Very long runs are split into focused sub-sessions with checkpoint saves, allowing the context window to reset
4. Profile-aware thresholds - Each profile has different compaction buffer sizes tuned to its expected conversation length

These optimizations reduce context exhaustion from approximately 15% of daily runs to under 2%.

Parallel Tool Execution

When AICOS identifies multiple independent tool calls that can run concurrently, it executes them in parallel rather than sequentially. This reduces wall-clock time for long processing runs by 30-50% for batches of independent operations like checking multiple project statuses or retrieving knowledge from several sources simultaneously.

Best Practices

Optimizing Execution

1. Set Realistic Goals - Overly ambitious goals lead to constant partial progress
2. Prioritize Clearly - Use goal and task priorities to guide AICOS's focus
3. Review Activity Logs - Monitor execution patterns to identify optimization opportunities
4. Respond to Approvals - Pending approvals block progress
5. Configure a Chat Model - Set a fast model for chat to get sub-second first-token responses
6. Monitor Profile Metrics - Use the Performance Dashboard to track costs and latency by profile

Managing Workload

1. Limit Active Goals - 3-5 active goals allows focused execution
2. Archive Completed Work - Keep the active list manageable
3. Break Down Large Tasks - Smaller tasks show better progress tracking
4. Set Reasonable Deadlines - Allow buffer for unexpected blockers

Related Topics

• AICOS Setup - Initial configuration
• Knowledge Management - How AICOS learns
• Tools Reference - Available actions
• Settings & Customization - Configuration options
• Execution Profiles - Profile definitions and customization
• Performance Dashboard - Monitor execution metrics and costs