curiositech/bdi-soft-systems

SKILL: Modeling Organizations as BDI Agents

When to Use This Skill

Load this skill when facing:

• The formalization gap: You have rich qualitative data about organizational behavior but need predictive models
• The black box problem: Existing process models describe WHAT happens but not HOW decisions get made internally
• Multi-agent coordination puzzles: Agents must work together despite different perceptions, goals, and expertise levels
• Implementation resistance: Technical systems fail because they don't match how organizations actually think and decide

Decision Points

1. Choosing Abstraction Level for Modeling

IF need to predict specific decisions → Micro-level (individual agents)
├─ Map each person/department as separate BDI agent
├─ Model conflicts and negotiations explicitly
└─ Use when: budget allocation, hiring decisions, project prioritization

IF need to understand process patterns → Meso-level (process as agent)
├─ Model entire workflow as single distributed agent
├─ Focus on recurring behaviors and bottlenecks
└─ Use when: IT development cycles, approval chains, quality control

IF need to explain culture/identity → Macro-level (organization as agent)
├─ Model whole organization as one meta-cognitive system
├─ Focus on identity, values, and strategic direction
└─ Use when: M&A integration, culture change, strategic pivots

2. Designing Agent Commitment Levels

Environment Stability × Information Cost Decision Matrix:

Stable Environment:
├─ Cheap Information → Moderate intentions (monthly review cycles)
└─ Expensive Information → Strong intentions (annual planning cycles)

Volatile Environment:
├─ Cheap Information → Weak intentions (daily standups, rapid iteration)
└─ Expensive Information → ESCALATE: Need higher-level coordination

IF coordination cost > decision complexity → Synchronize intention formation
IF agents have conflicting perceptual filters → Design accommodations, not consensus

3. Handling Agent Disagreements

Type of Disagreement → Diagnostic Check → Action

Factual Disagreement:
├─ Check: Different roles/expertise/positions?
├─ If YES → Map perceptual filters, design information accommodations
└─ If NO → Look for hidden goal conflicts

Goal Disagreement:
├─ Check: Has genuine negotiation occurred?
├─ If NO → Facilitate discourse process for desire formation
└─ If YES → Design accommodation mechanisms for persistent differences

Action Despite Disagreement:
├─ Check: Are there implicit accommodations?
├─ Surface and formalize working agreements
└─ Don't force consensus—optimize for coordination

Failure Modes

1. Consensus Paralysis

Symptoms: Endless meetings, recurring discussions, "we need alignment" language, delayed decisions waiting for agreement Detection Rule: If same decision requires 3+ consensus-building sessions, you're in consensus paralysis Fix: Switch to accommodation design—create working agreements that allow action despite disagreement

2. Intention Drift

Symptoms: Agents constantly reconsider commitments, plans change weekly, "keeping options open" becomes paralysis Detection Rule: If agents spend >30% of cycles reconsidering prior commitments, intentions are too weak Fix: Strengthen intention commitment thresholds; only reconsider on explicit trigger events

3. Perceptual Projection

Symptoms: Assuming others have access to same information, designing systems around "shared situational awareness" Detection Rule: If coordination failures blamed on "communication issues" without role/expertise analysis Fix: Map perceptual filters by role; design for subjective beliefs, not objective world state

4. Goal Optimization Fallacy

Symptoms: Treating organizational objectives as fixed inputs to optimize, missing how goals emerge from discourse Detection Rule: If designing for given objectives without modeling goal formation process Fix: Model the discourse mechanisms that create desires; include power dynamics and negotiation processes

5. System-Organization Conflation

Symptoms: Designing IT/AI systems as if they constitute organizational cognition, expecting radical behavior change from system deployment Detection Rule: If system failure would halt organizational function entirely (vs. reducing efficiency) Fix: Design systems as cognitive prosthetics that extend but don't replace organizational thinking

Worked Examples

Example 1: IT Department Resistance to New DevOps Platform

Scenario: CTO mandates transition to new DevOps platform. Development teams resist, claiming it "doesn't fit our workflow." Project stalls.

Novice Analysis: "They're just resistant to change. Need better training and change management."

Expert BDI Analysis:

1. Level Selection: Use micro-level (individual agent) modeling—need to predict specific adoption decisions
2. Perceptual Filter Mapping:
   • CTO perceives: strategic efficiency, vendor relationships, industry standards
   • Developers perceive: daily tool friction, debugging complexity, deployment reliability
3. Intention Analysis: CTO formed strong intention (annual planning cycle), developers have weak intentions (sprint-level)
4. Accommodation Design: Instead of forcing consensus, create accommodation—parallel toolchain during transition, developer input on configuration

Trade-offs Identified:

• CTO's strategic intention vs. developers' operational flexibility
• Platform standardization vs. workflow customization
• Short-term productivity loss vs. long-term efficiency gains

Resolution: Design accommodation allowing gradual migration with developer customization input, rather than forcing consensus on "best" approach.

Example 2: M&A Integration Failure

Scenario: Large corp acquires startup. 18 months later, startup talent has fled, innovation has stopped, integration deemed failure.

Expert BDI Analysis:

1. Level Selection: Macro-level—examining organizational identity and culture clash
2. Belief-Desire-Intention Mapping:
   • Large Corp Agent: Beliefs (process standardization works), Desires (predictable outcomes), Intentions (integrate within 12 months)
   • Startup Agent: Beliefs (agility requires flexibility), Desires (rapid innovation), Intentions (maintain autonomy)
3. Discourse Analysis: No genuine negotiation occurred—integration was mandated, not accommodated
4. Failure Mode: Consensus assumption—assumed shared goals without modeling desire formation process

What Novice Missed: Treated integration as operational problem rather than cognitive coordination challenge between incompatible agent architectures.

Expert Intervention: Design dual-agent accommodation—startup maintains separate BDI structure while creating interface protocols for coordination with parent org.

Quality Gates

• [ ] All key organizational agents identified at appropriate abstraction level
• [ ] Each agent's perceptual filters explicitly mapped (role, expertise, position-based)
• [ ] Intention formation and commitment thresholds specified for each agent type
• [ ] Disagreement accommodation mechanisms designed (not consensus requirements)
• [ ] Model predictions compared against actual organizational behavior patterns
• [ ] System interfaces aligned with organizational abstraction level boundaries
• [ ] Desire formation processes modeled (not just satisfaction of given goals)
• [ ] Information flow designed for subjective beliefs, not objective shared state
• [ ] Escalation triggers defined for when abstraction level coordination breaks down
• [ ] Failure mode detection rules testable against observable organizational symptoms

NOT-FOR Boundaries

This skill is NOT for:

• Simple process automation → Use standard workflow modeling instead
• Individual psychology modeling → Use cognitive psychology frameworks instead
• Technical system optimization → Use operations research methods instead
• Legal/compliance requirements → Use regulatory framework analysis instead
• Financial modeling → Use economic analysis frameworks instead

Delegate to other skills when:

• Need individual behavioral prediction → Use cognitive-behavioral-modeling
• Need technical performance optimization → Use system-performance-analysis
• Need regulatory compliance design → Use compliance-framework-design
• Need pure multi-agent coordination → Use distributed-systems-coordination
• Need organizational change management → Use change-management-strategy

This skill specifically addresses: The cognitive architecture of organizational decision-making where multiple agents with different expertise levels must coordinate through discourse and accommodation rather than consensus or authority.