In modern network environments, the exponential growth in data traffic and architectural complexity has made the centralized management of Software-Defined Networks (SDNs) increasingly challenging. Traditional SDN controllers, while programmable, still depend on static rules and human intervention for policy definition and threat mitigation, limiting their responsiveness in dynamic and large-scale deployments. This paper introduces a Large Language Model (LLM)-based agentic framework designed to enhance the cognitive and autonomous capabilities of SDN control planes. By integrating advanced reasoning and contextual understanding from LLMs, the proposed system enables intelligent policy generation, adaptive network reconfiguration, and explainable decision-making in real time.The framework leverages Mininet for virtual SDN simulation, Suricata as a telemetry and event-generation layer, and a multi-agent LLM pipeline for interpreting network events, classifying contextual significance, and dynamically generating flow rules. These rules are executed through an OpenDaylight SDN controller, allowing the network to respond automatically to anomalies, performance degradations, or security events without manual intervention. This closed-loop automation demonstrates how LLM-driven reasoning can transform SDN architectures from reactive, rule-based systems into proactive, self-optimizing infrastructures. The proposed approach supports continuous adaptation, scalability across cloud and IoT environments, and transparency through human-readable reports, marking a step toward fully autonomous and explainable network orchestration.

Collaborative robots promise agility and resilience in manufacturing, yet many deployments fail to deliver tangible relief for frontline employees. Addressing this gap, we present a transparent, LLM-supported procedure for “shift handover to the robot,” which systematically delegates repetitive, routine tasks to an unmanned robot shift while preserving human oversight and value creation during staffed hours. The framework operationalises ambidextrous automation: it protects the exploitation of proven processes while intentionally freeing human time and attention for exploration activities such as improvement, onboarding, observation, and learning. The central thesis is that transparent, data-grounded delegation can convert the potential of collaborative robotics into realised employee relief and stable productivity. This paper summarises context, approach, outcomes, and brief preliminary validation results of the procedure, including “exploration surplus” and practical SME adoption guidance.

At the core is an interactive, checklist-guided decision process that combines fourteen practical criteria with context data from production systems. Evidence includes cycle times, takt adherence, setup and changeover characteristics, error history, scheduling buffers, and environmental, health, and safety constraints. A retrieval-augmented LLM interprets this evidence and proposes candidate tasks for delegation, supporting both cooperative day-shift work and an autonomous, unmanned robot shift. Each recommendation carries a transparent decision trail – prompt-response pairs, data inputs, and scored criteria – that forms standardised handover documentation and enables auditability.

The procedure includes a lightweight task library, exception-handling rules (human fallback, safe stops, and alerts), and scheduling integration so that the robot shift is treated as a governed planning resource rather than ad-hoc automation. We anchor the design in ambidexterity theory by positioning exploration and exploitation as mutually reinforcing over time. Delegating monotony to the robot creates an “exploration surplus”: time and cognitive bandwidth that humans reinvest in improvement and innovation, which, in turn, yields more exploitable routines for subsequent robot shifts. This virtuous cycle is illustrated conceptually through an idealised productivity curve and can be tracked empirically using operational metrics.

We outline measurable targets for repeatable adoption in small and medium-sized enterprises: quantitative indicators include OEE deltas, interruption counts, idle time in unmanned shifts, error rates, and rework; qualitative indicators include perceived workload, trust, and collaboration quality. Expected outcomes are fewer operator interruptions, more stable after-hours execution, and higher acceptance due to clear, auditable decisions. Because the criteria and data interfaces are explicit, the approach is pragmatic to implement and portable across heterogeneous equipment bases. Limitations include data availability and quality, the need for robust safeguards in unmanned operation, and disciplined management of prompts and knowledge bases; we propose mitigations such as minimum data thresholds, watchdog mechanisms, and continuous review of decision logs.

Taken together, the procedure reframes robot handover from a narrow technical integration to a socio-technical management practice. It clarifies which tasks are suitable, why they are suitable, and how their delegation will be evaluated, making automation choices legible to engineers, planners, and operators alike. By turning the robot shift into an accountable planning instrument, the framework helps factories convert potential into realised benefits with traceability, safety, and human-centred relief at scale.

Throughout the entire life cycle of stationary gas turbine units (GTUs), such as industrial gas turbine power stations and gas pumping units, numerous types of tests are carried out. The results of these tests must be documented in reporting materials, including current test records with unit operating parameters over the course of testing, final consolidated protocols for each stage, summary data tables for operational documentation (e.g., passports, technical formulars, etc.), and other related records. The content and format of such documentation are defined by the manufacturer’s requirements and applicable standards, essentially representing standardized forms with a clear structure.In the work authors presenting the easy prompt-base solution for the automated test documentation generation. Implementation of the LLM for the test report generation is allowing to reduce the time for the developing working documentation and reduce the human error.

The automatic generation of Programmable Logic Controllers (PLCs) code serves as a tool for enhancing the intelligence and efficiency of manufacturing systems. PLCs can control a wide range of operations and processes, making them indis-pensable in today's industrial settings. One of the most significant innovations brought in this field of Artificial Intelligence (AI) is automated code generation.

The rapid advancement of AI has significantly impacted software develop-ment, particularly in automated code generation. AI-powered tools as machine learning and deep learning models aim to generate, optimize, and refine code, re-ducing development time and enhancing software quality. AI is becoming an in-telligent assistant, capable of accelerating software development phases. We first explore a comprehensive IEC 61131-3 international standard for multiple PLC programming languages as Structured Text (ST) and Ladder Diagram (LD). ST language is a high-level, text-based programming language used in PLC pro-gramming systems. Due to its structure and syntax, high-level programming lan-guages - like Delphi programming language - offer special readability and main-tainability, making it the preferred choice for developing complex control logic.

One main contribution of this paper is to design a structure for ST high-level programming language with its Delphi-like syntax. This is reflected in the case study where an application is presented that was first virtualized. And the pro-gram for the programmable logic controller that controls the operation of the mo-tors of a conveyor was created, in ST language, by using an AI generation tool that uses the natural programming language method.