Source-linked AI summary
An Autonomy Aware Metamodel for Human AI Collaboration in Software Engineering
Md Mahade Hasan, Md Toufique Hasan, Zheying Zhang, Kalle Kulonen, Konsta Kalliokoski, François Christophe, Tommi Mikkonen, Pekka Abrahamsson
TL;DR
Existing method engineering provides limited support for modeling authority that shifts across human–AI collaboration settings. This paper proposes a metamodel that derives autonomy from task, context, and collaboration pattern, then analytically illustrates dynamic authority assignment in requirements analysis.
Problem
Existing method engineering largely assumes static, human-centered authority and offers limited support for representing dynamic autonomy and situational authority redistribution.
Source: §1 · p. 1Method
The metamodel separates process structure from actor authority and models autonomy as situation-dependent across task, context, and collaboration pattern.
Source: §3 · p. 3, §3.1 · p. 3Results
The requirements-analysis instantiation illustrates autonomy variation across requirements generation and prioritization according to task responsibilities, team structure, decision model, and interaction rules.
Source: §6 · p. 7Takeaways & Limitations
The metamodel provides a conceptual foundation for software engineering methods that make authority distribution explicit and are adaptable and governable in human–AI collaboration.
Source: §5 · p. 6, §6 · p. 7As a vision paper, the work supports its proposal through an analytical instantiation with a multi-agent requirements analysis tool.
Source: §1 · p. 2
Abstract
from arXiv · showhide
Artificial Intelligence (AI) is shifting software engineering from tool-supported processes towards AI-first collaboration, where authority is dynamically distributed across human and artificial actors. However, existing method engineering approaches assume static, human-centric control and provide limited support explicitly capturing evolving autonomy. This paper presents a vision for autonomy-aware method engineering by proposing a metamodel that treats autonomy not as a fixed property of an actor, but as a derived, situation-dependent authority assignment determined by task, context, and collaboration pattern. The metamodel formalizes autonomy through four authority dimensions: task execution, task decomposition, task initiation, and collaboration reconfiguration. Through an analytical instantiation with a multi-agent requirements analysis tool, we illustrate how the metamodel supports dynamic authority assignment. This work provides a conceptual foundation for governance-aware, adaptable, and AI-first software engineering methods.
1 Introduction
AI-first software engineering shifts authority from stable human control toward dynamic distribution across human and artificial actors. The paper proposes an autonomy-aware metamodel and illustrates its use analytically in multi-agent requirements analysis.
1 Introduction
- AI agents increasingly execute tasks, decompose work, and coordinate activities, shifting software engineering toward dynamic authority distribution across actors and contexts.Earlier tool-assisted environments had more stable, human-centered authority.Source: §1 · p. 1
- Existing method engineering largely assumes static authority and does not explicitly model how autonomy varies across collaboration settings or governance constraints.The paper describes autonomy in multi-agent tools as typically implicit within execution models.Source: §1 · p. 1
- The proposed metamodel derives actor autonomy from task, context, and collaboration pattern while preserving stable process structure.It makes coordination and governance explicit through collaboration patterns and interaction rules.Source: §1 · pp. 1–2
- An analytical instantiation with a multi-agent requirements analysis tool illustrates variation in autonomy assignments and authority redistribution.The paper presents this as a conceptual foundation for adaptable software engineering methods in AI-first teams.Source: §1 · p. 2
2 Background and Related Work
Related work offers process metamodeling, ontology and agent-oriented approaches, and studies of human–AI collaboration, but does not unite method structure with dynamic authority redistribution. The paper positions autonomy-aware method modeling as a method-engineering concern.
2 Background and Related Work
- Process metamodels support reusable and analyzable method structures, while largely assuming static, human-centered authority assignments.SPEM 2.0, ISO/IEC 24744, and SEMAT Essence represent method elements such as roles, tasks, and work products.Source: §2 · p. 2
- Ontology and agent-oriented work models semantics and interactions, but does not provide a process metamodel combining method structure with authority distribution.Multi-agent systems and human–AI collaboration studies also raise coordination, transparency, and governance concerns [21] [9].Source: §2 · p. 2
- The proposed approach treats autonomy as a method-engineering construct for varying authority across actors without redesigning process structure.This contrasts with workflow-level views that do not specify authority redistribution as a method-modeling concern.Source: §2 · p. 2
3 Autonomy-Aware Metamodel Definition
The metamodel separates the stable structure of a human–AI software engineering method from situation-dependent authority assignment. It represents autonomy through four authority dimensions and five progressively broader levels.
3 Autonomy-Aware Metamodel Definition
- The metamodel represents autonomy as actor-level authority that varies with situation, independently of the method’s core process structure.It models human and artificial participants uniformly as actors within a shared socio-technical method structure.Source: §3 · p. 3
3.1 Core Constructs
- Seven core constructs represent actors, tasks, capabilities, artifacts, collaboration patterns, constraints, and context in a method.Constraints apply uniformly to actors, tasks, and artifacts through a common superclass.Source: §3.1 · p. 3
- Actors hold context-dependent authority over task execution, decomposition, initiation, and modification of collaboration structures.Tasks link required capabilities and artifacts and can be ordered or decomposed; capability-based allocation supports substitution between human and AI actors.Source: §3.1 · pp. 3–4
- Collaboration patterns specify interaction rules, team structure, and decision models, while context describes situational applicability and constraints prescribe permitted behavior.Artifacts provide traceability, coordination, and evaluation across tasks and actors.Source: §3.1 · p. 4
3.2 Autonomy Levels
- Autonomy depends on task, context, and collaboration pattern, so the same actor may receive different authority assignments across situations.The four dimensions are task execution, task decomposition, task initiation, and collaboration reconfiguration.Source: §3.2 · p. 4, §3.2.1 · p. 4
3.2.2 Autonomy Level Definition.
- Five autonomy levels, AL0 through AL4, describe progressively broader control, from a passive tool to an autonomous orchestrator.At AL3, actors may execute, decompose, and initiate tasks within scope; at AL4, they may also modify collaboration patterns, with human oversight shifting to supervision.Source: §3.2.2 · pp. 4–5
4 Illustrative Instantiation of the Metamodel
The requirements-analysis instantiation illustrates how collaboration patterns and decision mechanisms shape authority across two tasks. Autonomy varies by task responsibility and collaboration pattern, while the method structure can remain stable.
4 Illustrative Instantiation of the Metamodel
- Figure 2 depicts interactions among the human and PO, Dev, QA, and Manager agents, their tasks and artifacts, and the generation and prioritization scenarios.The diagram omits context and constraint instantiations and instance attributes, which are available in the repository.Source: §4 · p. 5, Figure 2 · p. 6
- Requirements generation uses hierarchical collaboration with HumanApproval, while prioritization uses centralized voting, with the Manager aggregating agent lists and resolving conflicts according to the User vote.Generation delegates through the PO, which coordinates Dev and QA suggestions before returning requirements for human approval.Source: §4 · p. 5, Table 2 · p. 6
- During requirements generation, the PO operates at AL2 while Dev and QA operate at AL1; during prioritization, Dev and QA rise to AL2 by generating lists that contribute directly to the outcome.The same requirements-engineering context supports different autonomy assignments as task responsibilities and collaboration patterns change.Source: §4 · p. 5
- The instantiation presents mechanisms for modeling collaboration, decision-making, and authority redistribution, allowing the same method to evolve toward more autonomous AI teamwork without changing its underlying process structure.Source: §4 · p. 5
5 Discussion and Future Plans
The metamodel separates stable method structure from authority assignment, treating autonomy as situation-dependent. The discussion frames it as an initial step and identifies four research directions toward an empirically grounded, tool-supported approach.
5 Discussion and Future Plans
- The metamodel separates stable representations of method elements from autonomy levels that vary with task, context, and collaboration pattern, with governance conditions and decision arrangements shaping authority rather than technical capability alone.The paper presents this as an initial step toward adaptable methods that preserve appropriate human oversight and governance.Source: §5 · p. 6
- Future work will refine the autonomy function by identifying how task, context, constraint, collaboration-pattern, team, decision, interaction, criticality, and governance parameters affect permissible autonomy.Source: §5 · p. 6
- Comparative case studies across requirements analysis, architecture design, coding, testing, and maintenance will assess whether the constructs represent diverse settings or need additions.Source: §5 · p. 6
- Planned AI-assisted tooling could extract method elements from process descriptions, traces, or scenarios, suggest autonomy levels, and check governance consistency under human review.The tool-support proposal is described across the passage introducing it and its continuation.Source: §5 · pp. 6–7
- Future work will investigate runtime adaptation using monitoring to change authority allocation with task criticality, uncertainty, regulatory conditions, performance, or human trust.The aim is to extend the metamodel from design-time specification toward a governance mechanism for evolving teams and, ultimately, an empirically grounded, tool-supported approach.Source: §5 · p. 7
6 Conclusion
The conclusion presents autonomy as authority derived from task, context, and collaboration pattern, formalized across four authority dimensions. An illustrative tool instantiation shows variation across requirements tasks and supports a conceptual foundation for adaptable, governable methods.
6 Conclusion
- The metamodel defines autonomy as situation-dependent authority over task execution, task decomposition, task initiation, and collaboration reconfiguration.Separating method structure from authority assignment lets engineers model stable collaboration structures while analyzing authority distribution among human and AI actors.Source: §6 · p. 7
- The multi-agent requirements-analysis instantiation shows autonomy varying across requirements generation and prioritization according to task responsibilities, team structure, decision model, and interaction rules.Source: §6 · p. 7
- The work offers a conceptual foundation for software-engineering methods that are governable, adaptable, and explicit about authority distribution in human–AI collaboration.Source: §6 · p. 7