Abstract
Note. This article presents a conceptual hypothesis intended as the foundation for further exploration. It serves as a conceptual model that will be refined through continued engineering practice, case studies, and comparison with existing research.
While current AI Engineering primarily focuses on building increasingly capable AI systems, this article approaches the problem from a different perspective. Rather than asking how AI assistants should be implemented, it asks how they should be specified when performing organizational work.
This conceptual contribution argues that AI assistants should be specified through organizational job roles rather than as isolated software components. The central hypothesis is that the same organizational expectations that define the work of a human specialist can also serve as the specification for an AI assistant.
Rather than prescribing detailed workflows, a Job Role Specification describes the purpose of a role, its responsibilities, expected inputs and outputs, quality expectations, constraints, success criteria, and guiding principles. The realization of this specification—whether through a human specialist, traditional software, an AI assistant, or a hybrid collaboration—becomes a separate engineering concern.
The ideas presented in this article originated from the engineering of an AI-supported publication workflow. During this work, individual AI assistants gradually evolved into specialized editorial roles whose behavior was defined less by prompts than by clearly specified organizational responsibilities. This observation motivated the conceptual model presented in this article and forms the foundation for its continued development.
1. Introduction
AI assistants are increasingly used to summarize documents, analyze information, generate reports, and review artifacts.
Current AI Engineering primarily concentrates on the implementation of these assistants. Research focuses on prompts, reasoning strategies, memory architectures, agent orchestration, tool use, planning, and multi-agent collaboration.
This article considers a preceding engineering question.
Rather than asking how AI assistants should be implemented, it asks how they should be specified when they perform organizational work.
Human specialists are typically not guided through detailed procedural instructions. Instead, organizations define responsibilities, expected work results, quality expectations, collaboration with other organizational roles, and organizational objectives. Much of the remaining knowledge is implicit, acquired through education, professional experience, organizational culture, and continuous learning.
This implicit organizational understanding enables specialists to interpret tasks, resolve ambiguity, and make informed decisions without requiring exhaustive procedural specifications.
This article proposes that the same principle can serve as the primary engineering artifact for AI assistants.
Consequently, whenever AI assistants are intended to perform organizational work and collaborate with other organizational actors, they should first be specified as organizational job roles. Their technical realization becomes a subsequent engineering activity.
This perspective intentionally separates organizational specification from technical realization.
2. Organizational Architecture
The proposed model distinguishes three complementary architectural perspectives.
Three concepts are used consistently throughout this article.
- Capability describes what an organization needs in order to accomplish its objectives.
- Job Role describes an organizational responsibility together with its purpose, expected work results, quality expectations, constraints, success criteria, and organizational interfaces.
- Realization describes how a Job Role Specification is realized.
The key observation underlying this model is straightforward:
Technologies evolve continuously, whereas organizational responsibilities evolve much more slowly.
Organizations therefore benefit from specifying stable organizational responsibilities independently of the technologies used to realize them.
Business View
The Business View defines what the organization intends to accomplish. It identifies the capabilities required to achieve organizational objectives without making assumptions about realization.
Organizational View
The Organizational View defines how organizational work is structured.
Its central element is the Job Role. A job role represents an organizational responsibility together with its expected work results. It is an organizational abstraction rather than a person or a technical system.
During the engineering of an AI-supported publication workflow, individual AI assistants gradually evolved into specialized editorial roles. They became distinguished less by prompts than by clearly defined responsibilities, expected work results, and collaboration with other assistants.
This observation suggests that AI assistants should first be specified as organizational job roles before realization decisions are made.
Realization View
The Realization View describes how a Job Role Specification is realized.
Possible realizations include human specialists, traditional software, AI assistants, or hybrid human–AI collaboration.
The organizational specification remains stable while realization technologies evolve.
Relationship Between the Views
Business View
Capabilities
──────────────────────────────
Organizational View
Job Roles
│
▼
Work Results
──────────────────────────────
Realization View
Human Specialist
Traditional Software
AI Assistant
Hybrid Collaboration
The Business View defines what the organization requires.
The Organizational View defines what is expected.
The Realization View defines how these expectations are realized.
3. From Organizational Knowledge to Job Role Specifications
Human specialists continuously draw upon implicit organizational knowledge acquired through education, professional experience, organizational culture, and everyday collaboration.
AI assistants cannot rely on this implicit knowledge.
Consequently, the organizational expectations associated with a role must be specified explicitly before they can be realized by an AI assistant.
This article proposes the Job Role Specification as the primary engineering artifact for this purpose.
Rather than describing implementation details, a Job Role Specification captures the organizational expectations associated with a role. It defines its purpose, responsibilities, expected inputs and outputs, organizational interfaces, constraints, success criteria, quality expectations, and guiding principles.
Its objective is not to prescribe every individual action. Instead, it communicates the organizational intent of the role while leaving sufficient autonomy for its realization.
This distinction emerged during the engineering of AI-supported publication workflows.
Increasingly detailed procedural instructions consistently reduced the quality of generated results. As specifications became more elaborate, they also became increasingly difficult to maintain while limiting the assistant’s ability to reason about unfamiliar situations.
In contrast, specifications centered on organizational purpose, responsibilities, success criteria, quality expectations, and guiding principles consistently produced more robust behavior.
This resembles how organizations guide experienced specialists. Rather than prescribing every individual action, organizations communicate responsibilities, expected work results, and organizational principles that enable specialists to make informed decisions within their area of responsibility.
The same principle is proposed for AI assistants.
The Job Role Specification therefore becomes a stable organizational artifact from which AI assistants can be realized while preserving consistent organizational expectations.
4. Design Principle: Specify Intent, Not Execution
A Job Role Specification should describe what a role is expected to accomplish, not how every individual task should be executed.
Its purpose is to establish a clear organizational contract rather than an executable procedure.
Accordingly, a Job Role Specification defines the purpose of a role, its responsibilities, expected work results, organizational interfaces, quality expectations, constraints, success criteria, and guiding principles.
These elements communicate organizational intent while leaving realization decisions to the assigned specialist, software system, or AI assistant.
An AI assistant should not merely follow instructions.
It should understand the objective of its role, the quality expected from its work, the organizational context in which it operates, and the principles that guide its decisions.
The objective is therefore not to minimize an assistant’s decision-making, but to constrain it through organizational intent rather than procedural instructions.
The proposed design principle can be summarized as follows:
Specify intent, not execution.
5. Outlook
This article introduces a conceptual hypothesis for specifying AI assistants through organizational job roles.
Its purpose is not to present a complete engineering methodology, but to establish a conceptual foundation for subsequent work.
The ideas presented here emerged from continued engineering practice during the development of an AI-supported publication workflow and are currently being transferred into the implementation of the Charlie platform, where Job Role Specifications serve as the foundation for executable AI assistants.
Future work will elaborate this concept by deriving a systematic specification model, identifying reusable design principles, and evaluating the approach through continued engineering practice and case studies.
The long-term objective is to establish Job Role Specifications as a reusable engineering artifact for designing AI assistants that operate within organizational environments.
