Liquid Interfaces: A Dynamic Ontology for the Interoperability of Autonomous Systems

Dhiogo José Correa de Sá, Carlos Pereira Lopes Filho, Carlos Eduardo Schmiedel · 2026

Read Liquid Interfaces: A Dynamic Ontology for the Interoperability of Autonomous Systems by Dhiogo José Correa de Sá, Carlos Pereira Lopes Filho, Carlos Eduardo Schmiedel (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2601-21993.txt.

Introduction

Dhiogo José Correa de Sá, Carlos Pereira Lopes Filho. And Carlos Eduardo Schmiedel argue that software interfaces are too rigid for the age of autonomous agents. Agents reason in adaptive, probabilistic, context-dependent ways. System integration, by contrast, still rests on static interfaces and fixed contracts between programs.

Their 2026 preprint introduces Liquid Interfaces, a coordination paradigm where an interface is not a lasting technical artifact. Instead, interfaces are treated as temporary, formed for the interaction at hand. The idea reframes integration as negotiation rather than plumbing. If it works, systems of agents could connect, cooperate, and part without the brittle hand-built adapters that make today's integrations so costly to maintain.

Catalog record

CREATORDhiogo José Correa de Sá, Carlos Pereira Lopes Filho, Carlos Eduardo Schmiedel
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE53 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2601-21993

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Liquid Interfaces: A Dynamic Ontology for the Interoperability of Autonomous Systems Dhiogo José Correa de Sá, Carlos Pereira Lopes Filho, Carlos Eduardo Schmiedel Abstract Contemporary software architectures struggle to support autonomous agents whose reasoning is adaptive, probabilistic, and context-dependent, while system integration remains dominated by static interfaces and deterministic contracts. This paper introduces Liquid Interfaces, a coordination paradigm in which interfaces are not persistent technical artifacts, but ephemeral relational events that emerge through intention articulation and semantic negotiation at runtime. We formalize this model and present the Liquid Interface Protocol (LIP), which governs intention-driven interaction, negotiated execution, and enforced ephemerality under semantic uncertainty. We further discuss the governance implications of this approach and describe a reference architecture that demonstrates practical feasibility. Liquid Interfaces provide a principled foundation for adaptive coordination in agent-based systems. Keywords: Liquid Interfaces; Intention Based Computing; autonomous agents; Semantic Interoperability; Software Architecture; Communication Protocols. 1 Introduction The history of system integration is fundamentally a chronicle of attempts to impose order upon informational chaos. From nascent Remote Procedure Call (RPC) protocols 4 to the ubiquity of REST 6 and GraphQL 9 , the overarching objective has been standardization. The foundational premise of traditional Software Engineering posits that security and efficiency are contingent upon predictability: System A must determine, with byte-level precision, what System B anticipates receiving. This paradigm, codified by Fielding’s architectural constraints 6 , has succeeded for decades, enabling the construction of robust, complex systems through explicit and immutable contracts. However, the integration of Large Language Models (LLMs) and Autonomous Agents into enterprise workflows has precipitated a fundamental ontological dissonance 25 . On one spectrum lies the liquid agent, which operates probabilistically, interprets natural language, adapts to ambiguous contexts, and pursues high-level objectives (e.g., “reduce logistics costs by 15 \% while maintaining delivery quality”) 23 . On the other lies the “solid API,” which operates deterministically, demands rigid syntax, and executes discrete functions (e.g., POST /orders/v1/create with a predefined JSON schema). This friction is not merely technical; it is ontological. As highlighted by recent research on tool-augmented language models, the capability of agents to interact with external tools is often constrained by the rigid structural requirements of existing interfaces 20 . Consider a practical scenario: an AI agent within a logistics firm must respond to a storm that has closed a port. The agent perceives the disruption, yet the legacy routing system’s API merely returns a static error: “Route Unavailable.” The agent cannot negotiate alternatives; it lacks the protocol to converse with disparate systems to synthesize a solution. It remains trapped in a deterministic impasse. Consequently, a human operator must intervene, manually querying siloed systems, negotiating with suppliers, and updating multiple platforms. What should be an autonomous decision devolves into a costly, manual process. The cost of translation between the agent’s fluid intent” and the API’s rigid form” has emerged as the primary bottleneck for intelligent automation. The maintenance of connectors, the synchronization of documentation (e.g., Swagger/OpenAPI), and the management of breaking changes are symptomatic of a paradigm that has reached its elastic limit. As noted in the context of machine learning systems, these integration glues and schema dependencies contribute significantly to hidden technical debt, compounding exponentially with each new system addition 21 . To transcend these limitations, we propose the Liquid Interface : a framework that eschews static form as a prerequisite for interaction, privileging flow and intent. Rather than rigid systems connected via static contracts, we propose a paradigm wherein interfaces emerge dynamically, negotiated at runtime between agents that may possess no prior history of interaction. A Liquid Interface is not an endpoint, a contract, or a schema. It is a relational event that manifests when two or more agents require collaboration. It arises from necessity, stabilizes only for the required duration, and dissolves once its purpose is fulfilled, leaving zero technical debt. In this model, the logistics agent could express its intent: “I need to deliver this container within a 200km radius of the closed port, optimizing for cost, time, and delivery quality.” The system would autonomously discover which agents (warehouses, carriers, suppliers) could assist, negotiate terms, and execute the solution—all without human intervention or pre-written integration code. This paper intentionally focuses on the conceptual, formal, and architectural foundations of Liquid Interfaces. Concrete implementations, performance trade-offs, and empirical evaluations are treated as separate concerns and are the subject of ongoing and future work. 2 Theoretical Foundation and Related Work Liquid Interfaces are proposed as an architectural response to the increasing heterogeneity, uncertainty, and volatility found in modern distributed computational environments. While existing paradigms provide robust abstractions for integrating well-defined services with stable contracts, they struggle in settings where participants are unknown at design-time, capabilities evolve dynamically, and meaning must be negotiated during execution. In such settings, static agreement mechanisms become sources of friction rather than enablers of coordination. 2.1 Architectural Motivation Most contemporary distributed systems assume that interacting components can establish a stable representational contract prior to execution. Microservices, Service Meshes, and API Gateways operate under this assumption by enforcing static schemas, predefined endpoints, and lifecycle policies that support syntactic interoperability. Semantic Web technologies extend this premise into the semantic domain by attempting to align vocabularies through global ontologies and shared taxonomies. Agent Communication Languages (ACLs) further assume that interacting agents share a common logical framework and can interpret performative acts within that logic. These architectures are effective when the system is closed, the interfaces are known in advance, and the domain is sufficiently stable. However, their design assumptions introduce friction in open-ended computational ecosystems in which (i) novel services may appear without prior registration, (ii) capabilities may vary over time due to learning or context, and (iii) goals may be situational rather than predetermined. In these environments, static schemas and global ontologies impose upfront alignment costs that scale poorly and reduce the flexibility of the agentic ecosystem. 2.2 Limits of LLM-Based Agentic Systems Recent research demonstrates that Large Language Models (LLMs) can autonomously plan, invoke tools, and orchestrate interactions across heterogeneous systems, effectively acting as general-purpose policy engines. While this represents a major advance in agentic architectures, scaling LLMs does not eliminate several intrinsic limitations. Recent theoretical work formalizes five such constraints 16 , namely: (i) inevitable hallucination due to inherent approximation error, (ii) context compression limitations that degrade reasoning over long inputs, (iii) brittle reasoning emerging from likelihood-based training, (iv) retrieval fragility caused by semantic mismatch between evidence and generation, and (v) multimodal misalignment that undermines robust cross-signal grounding. Empirical studies corroborate several of these phenomena: LLMs exhibit irreducible uncertainty even when confident 10 , and their ability to exploit long contexts deteriorates as distance from the query increases 14 . These limitations have two consequences for open agentic ecosystems. First, they imply that agentic systems cannot rely on deterministic correctness or complete semantic grounding, even when augmented with planning or retrieval mechanisms. Second, they indicate that LLMs can negotiate actions, but lack mechanisms for negotiating interfaces or meaning under ambiguity, especially when semantic context is asymmetrically distributed among participants. In other words, while LLMs can decide what to do, they struggle to decide how to talk about what is being done when no prior agreement exists. 2.3 Proposal: Liquid Interfaces The

Read the complete text