Bradshaw's (1997) synthesis of nine properties that define software agents — the foundational checklist that every modern "agentic" system can be evaluated against. ## Core Concept Drawing from IBM (Gilbert et al. 1995), Wooldridge and Jennings (1995), Franklin and Graesser (1996), Nwana (1996), and others, Bradshaw identified nine properties that researchers consistently use to characterize software agents. No single agent possesses all nine — they form a design space, not a checklist of requirements. ## The Nine Properties | # | Property | Definition | Modern Example (2026) | |---|----------|-----------|----------------------| | 1 | **Reactivity** | Ability to selectively sense and act on environmental changes | Claude Code reading file changes, test failures | | 2 | **Autonomy** | Goal-directedness, proactive and self-starting behavior | Agent loops that plan and execute without prompting | | 3 | **Collaborative behavior** | Working in concert with other agents toward common goals | Multi-agent orchestration (LangGraph, CrewAI) | | 4 | **Knowledge-level communication** | Ability to communicate using language resembling human discourse, not just protocol-level messages | Natural language tool calls, MCP protocol | | 5 | **Inferential capability** | Acting on abstract task specifications using prior knowledge, not just following literal instructions | LLMs reasoning about intent vs stated instructions | | 6 | **Temporal continuity** | Persistence of identity and state over long periods of time | Agent memory systems, conversation history | | 7 | **Personality** | Maintaining attributes of a "believable" character, such as emotion | Claude's communication style, system prompts | | 8 | **Adaptivity** | Ability to learn and improve with experience | Fine-tuning, RLHF, in-context learning | | 9 | **Mobility** | Ability to migrate in a self-directed way from one host to another | Telescript (1997), mobile code, serverless functions | ## The Three Primary Dimensions (IBM) Gilbert et al. (1995) proposed that all agents exist in a space defined by three dimensions: ``` Intelligence / / /_________ Agency (Autonomy) | | Mobility ``` - **Agency**: Degree of autonomy and authority vested in the agent - **Intelligence**: Ability to reason, learn, and carry out tasks adaptively - **Mobility**: Degree to which agents travel through networks ## Nwana's Typology (1996) Based on three primary attributes — **autonomy**, **cooperation**, **learning** — Nwana derived agent types: - **Smart Agents**: All three (autonomy + cooperation + learning) - **Collaborative Learning Agents**: Cooperation + learning - **Interface Agents**: Autonomy + learning - **Collaborative Agents**: Autonomy + cooperation - **Autonomous Agents**: Autonomy only - **Learning Agents**: Learning only ## Why This Framework Endures The nine properties provide a precise vocabulary for the imprecise term "agentic": 1. **Evaluation tool**: Score any system against the nine properties to assess actual agent capabilities vs marketing claims 2. **Design guide**: Identify which properties matter for your use case (not all nine are needed) 3. **Gap analysis**: Identify what's missing — e.g., most 2026 AI agents score high on reactivity, autonomy, inference, and communication but low on temporal continuity and mobility 4. **Hype filter**: When someone says "our product is agentic," ask which of the nine properties it actually has ### Modern Agent Scoring Example | Property | Claude Code | Traditional Cron Job | Self-Driving Car | |----------|------------|---------------------|-----------------| | Reactivity | High | Low (scheduled) | High | | Autonomy | Medium (permission gates) | High (runs unsupervised) | High | | Collaboration | Medium (MCP tools) | None | Medium (V2X) | | Communication | High (natural language) | None | Low (signals) | | Inference | High | None | High | | Temporal continuity | Low (session-bound) | High (persistent) | High | | Personality | Medium (system prompt) | None | None | | Adaptivity | Medium (in-context) | None | High (ML models) | | Mobility | None | None | High (physical) | ## Cross-Domain Applications - **AI-Assisted Development** — Provides the vocabulary to distinguish genuinely agentic coding tools from glorified autocomplete. A tool with only reactivity and inference isn't an agent — it's a smart responder. - **Organizational design** — Human teams also have these properties: a good team is reactive, autonomous, collaborative, communicative, inferential, persistent, has culture (personality), learns (adaptive), and can be redeployed (mobile). The framework evaluates organizational health. - **Knowledge management** — A Second Brain scores surprisingly well: reactive (responds to queries), persistent (temporal continuity), adaptive (grows with use), communicative (wiki-links). Low on autonomy and mobility. - **Hiring and evaluation** — Individual contributors can be assessed on the same dimensions: How reactive are they to changing requirements? How autonomous? How collaborative? The framework transcends software. ## Related Concepts - [[Agent Definition Duality]] — The philosophical foundation: are these properties *ascribed* or *described*? - [[Behavior Based Agent Architecture]] — Maes' approach prioritized reactivity and adaptivity over inference and communication - [[Intelligence Augmentation Paradigm]] — The IA thesis adds a meta-property: does the agent's autonomy augment or erode human capability? - [[Agent Loop Maturity Spectrum]] — Modern maturity levels roughly correspond to accumulating more of the nine properties - [[Agentic Engineering Definition]] — Modern definitional attempts that implicitly reference subsets of this framework ## Sources - Bradshaw, J.M. (1997). "An Introduction to Software Agents." Chapter 1 in *Software Agents*, AAAI Press/MIT Press, pp. 3-46. - Gilbert, D., Aparicio, M., Atkinson, B., et al. (1995). "IBM Intelligent Agent Strategy." IBM white paper. - Nwana, H. (1996). "Software Agents: An Overview." *Knowledge Engineering Review*, 11(3), 205-244. - Franklin, S. and Graesser, A. (1996). "Is it an Agent, or just a Program?" *Proceedings of the Third International Workshop on Agent Theories, Architectures, and Languages*. - [[Software Agents (book)]] — Source document with full chapter screenshots