The Agentic Paradigm faces a significant Software Engineering Absence, yielding Agentic systems commonly lacking robustness, observability, and evolvability. To address these deficiencies, we propose a principled engineering framework comprising Runtime Goal Refinement (RGR), Observable Cognitive Architecture (OCA), and Evolutionary Memory Architecture (EMA). In this framework, RGR ensures robustness and intent alignment via knowledge-constrained refinement and human-in-the-loop clarification; OCA builds an observable and maintainable white-box architecture using component decoupling, logic layering, and state-control separation; and EMA employs an execution-evolution dual-loop for evolvability. We implemented and empirically validated Fairy, a mobile GUI agent based on this framework. On RealMobile-Eval, our novel benchmark for ambiguous and complex tasks, Fairy outperformed the best SoTA baseline in user requirement completion by 33.7%. Subsequent controlled experiments, human-subject studies, and ablation studies further confirmed that the RGR enhances refinement accuracy and prevents intent deviation; the OCA improves maintainability; and the EMA is crucial for long-term performance. This research provides empirically validated specifications and a practical blueprint for building reliable, observable, and evolvable Agentic AI systems.

In mental health counseling, condensing dialogues into concise and relevant summaries (aka counseling notes) holds pivotal significance. Large Language Models (LLMs) exhibit remarkable capabilities in various generative tasks; however, their adaptation to domain-specific intricacies remains challenging, especially within mental health contexts. Unlike standard LLMs, mental health experts first plan to apply domain knowledge in writing summaries. Our work enhances LLMs' ability by introducing a novel planning engine to orchestrate structuring knowledge alignment. To achieve high-order planning, we divide knowledge encapsulation into two major phases: (i) holding dialogue structure and (ii) incorporating domain-specific knowledge. We employ a planning engine on Llama-2, resulting in a novel framework, PIECE. Our proposed system employs knowledge filtering-cum-scaffolding to encapsulate domain knowledge. Additionally, PIECE leverages sheaf convolution learning to enhance its understanding of the dialogue's structural nuances. We compare PIECE with 14 baseline methods and observe a significant improvement across ROUGE and Bleurt scores. Further, expert evaluation and analyses validate the generation quality to be effective, sometimes even surpassing the gold standard. We further benchmark PIECE with other LLMs and report improvement, including Llama-2 (+2.72%), Mistral (+2.04%), and Zephyr (+1.59%), to justify the generalizability of the planning engine.

Automated planning is a prominent area of Artificial Intelligence, and an important component for intelligent autonomous agents. A cornerstone of domain-independent planning is the separation between planning logic, i.e. the automated reasoning side, and the knowledge model, that encodes a formal representation of domain knowledge needed to reason upon a given problem to synthesise a solution plan. Such a separation enables the use of reformulation techniques, which transform how a model is represented in order to improve the efficiency of plan generation. Over the past decades, significant research effort has been devoted to the design of reformulation techniques. In this paper, we present a systematic review of the large body of work on reformulation techniques for classical planning, aiming to provide a holistic view of the field and to foster future research in the area. As a tangible outcome, we provide a qualitative comparison of the existing classes of techniques, that can help researchers gain an overview of their strengths and weaknesses.

PDDL+ models are advanced models of hybrid systems and the resulting problems are notoriously difficult for planning engines to cope with. An additional limiting factor for the exploitation of PDDL+ approaches in real-world applications is the restricted number of domain-independent planning engines that can reason upon those models. With the aim of deepening the understanding of PDDL+ models, in this work, we study a novel mapping between a time discretisation of pddl+ and numeric planning as for PDDL2.1 (level 2). The proposed mapping not only clarifies the relationship between these two formalisms but also enables the use of a wider pool of engines, thus fostering the use of hybrid planning in real-world applications. Our experimental analysis shows the usefulness of the proposed translation and demonstrates the potential of the approach for improving the solvability of complex PDDL+ instances.

Despite all this promise, adoption of AI is not where many expected (or hoped) that it might be. Research continues to improve the underlying AI technology--the recent of development of both Midjourney and Stable Diffusion4 is a case in point--and firms continue to invest in AI.5 We even saw a bump in investment during the first couple of years of the pandemic.6 However, a majority of AI projects fail.7 Compelling demonstrations are not transitioning into value creating solutions. Autonomous cars are a prime example, where commercial, mass market, versions constantly seem to be a decade away, despite early success and significant investment. We hear a similar story from AI practitioners working in firms attempting to leverage AI, with carefully developed models and solutions left on the bench as they are either not compelling enough or too fragile to replace existing solutions. There are notable successes, such as machine language translation, however there appears to have been more misses.

Macro-operators (macros) are a well-known technique for enhancing performance of planning engines by providing "short-cuts" in the state space. Existing macro learning systems usually generate macros by considering most frequent action sequences in training plans. Unfortunately, frequent action sequences might not capture meaningful activities as a whole, leading to a limited beneficial impact for the planning process. In this paper, inspired by resource locking in critical sections in parallel computing, we propose a technique that generates macros able to capture whole activities in which limited resources (e.g., a robotic hand, or a truck) are used. Specifically, such a Critical Section macro starts by locking the resource (e.g., grabbing an object), continues by using the resource (e.g., manipulating the object) and finishes by releasing the resource (e.g., dropping the object). Hence, such a macro bridges states in which the resource is locked and cannot be used. We also introduce versions of Critical Section macros dealing with multiple resources and phased locks. Usefulness of macros is evaluated using a range of state-of-the-art planners, and a large number of benchmarks from the deterministic and learning tracks of recent editions of the International Planning Competition.

In Automated Planning, learning and exploiting additional knowledge within a domain model, in order to improve plan generation speed-up and increase the scope of problems solved, has attracted much research. Reformulation techniques such as those based on macro-operators or entanglements are very promising because they are to some extent domain model and planning engine independent. This paper aims to exploit recent work on inner entanglements, relations between pairs of planning operators and predicates encapsulating exclusivity of predicate achievements or requirements', for generating macro-operators. We discuss conditions which are necessary for generating such macro-operators and conditions that allow removing primitive operators without compromising solvability of a given (class of) problem(s). The effectiveness of our approach will be experimentally shown on a set of well-known benchmark domains using several high-performing planning engines.

Domain-independent planning requires only to specify planning problems in a standard language (e.g.

Capturing and exploiting structural knowledge of planning problems has shown to be a successful strategy for making the planning process more efficient. Plans can be decomposed into its constituent coherent subplans, called blocks, that encapsulate some effects and preconditions, reducing interference and thus allowing more deordering of plans. According to the nature of blocks, they can be straightforwardly transformed into useful macro-operators (shortly, macros). Macros are well known and widely studied kind of structural knowledge because they can be easily encoded in the domain model and thus exploited by standard planning engines. In this paper, we introduce a method, called BloMa, that learns domain-specific macros from plans, decomposed into macro-blocks which are extensions of blocks, utilising structural knowledge they capture. In contrast to existing macro learning techniques, macro-blocks are often able to capture high-level activities that form a basis for useful longer macros (i.e.

The development of domain-independent planners within the AI Planning community is leading to "off-the-shelf" technology that can be used in a wide range of applications. Moreover, it allows a modular approach --in which planners and domain knowledge are modules of larger software applications-- that facilitates substitutions or improvements of individual modules without changing the rest of the system. This approach also supports the use of reformulation and configuration techniques, which transform how a model is represented in order to improve the efficiency of plan generation. In this article, we investigate how the performance of domain-independent planners is affected by domain model configuration, i.e., the order in which elements are ordered in the model, particularly in the light of planner comparisons. We then introduce techniques for the online and offline configuration of domain models, and we analyse the impact of domain model configuration on other reformulation approaches, such as macros.