Pith. sign in

REVIEW 4 major objections 6 minor 2 cited by

The (R)evolution of Scientific Workflows in the Agentic AI Era: Towards Autonomous Science

T0 review · 4 major / 6 minor · reviewed 2026-08-04 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Scientific workflows and AI agents share one state-machine core, so autonomous science can be reached by gradual upgrades along two axes.

desk verdict A genuinely useful state-machine unification of workflows and agents, but the 5x5 evolution matrix is a narrative device, not a demonstrated structure, and the 100x figure is imported from the authors' own roadmap. read the letter →

arxiv 2509.09915 v1 pith:WZIOTJRE submitted 2025-09-12 cs.AI cs.DC

Woong Shin , Renan Souza , Daniel Rosendo , Frédéric Suter , Feiyi Wang , Prasanna Balaprakash , Rafael Ferreira da Silva This is my paper
classification cs.AIcs.DC
keywords scientificworkflowsagenticAIautonomoussciencestatemachineabstractionworkflowevolutionswarmintelligencemulti-agentsystemsmanagement
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper argues that the workflow systems used in science and the new agentic AI systems are not separate paradigms: both can be represented as state machines, an agent being anything that perceives and acts. From that common foundation, the authors define two independent evolutionary dimensions — intelligence (static to adaptive to learning to optimizing to intelligent) and composition (single to pipeline to hierarchical to mesh to swarm) — and place every scientific system on a 5×5 grid. The grid works as a taxonomy for existing systems and as a roadmap: moving from today's DAG-based pipelines toward fully autonomous, distributed laboratories is a sequence of small, optional upgrades rather than a disruptive rewrite. If the framework holds, the community can preserve two decades of workflow infrastructure while steadily adding reasoning, goal-seeking, and swarm coordination, with discovery acceleration of 10–100× as the payoff.

What carries the argument

The state machine M=(S,Σ,δ,s0,F) is the common denominator: workflow DAGs map directly onto it, and AI agents are the same machine with enhanced transition functions. Sophistication of δ defines the intelligence dimension — static δ:S×Σ→S, adaptive δ:S×Σ×O→S, learning via δ_{t+1}=L(δ_t,H), optimizing via δ*=argmin_δ J(δ), and intelligent via M'=Ω(M,C,G) — while composition operators (single M; pipeline M1∘M2∘...; hierarchical M_mgr(M1,...,Mn); mesh Mi↔Mj; swarm M=Φ({m1,...,mn})) define the composition dimension. The 5×5 matrix that results is the framework's engine: it maps any system to a cell and makes an upgrade path explicit.

What would settle it

Take any deployed scientific system and test the orthogonality prediction: an ML-guided workflow with a Learning-level transition function but organized as a single isolated machine (no pipeline or hierarchy) should be a valid, stable cell on the grid. If such a system cannot function without at least Pipeline composition, or if a Static system can be composed as a Swarm without intermediate levels, the dimensional independence breaks. More directly, find a real autonomous laboratory that reached the Intelligent×Swarm cell without passing through the intermediate intelligence and composition l

Watch

Extended reading notes

Core claim

The central claim is that autonomy in both scientific workflows and AI agents reduces to a single primitive — the state machine — whose transition function can be made progressively more sophisticated (adding feedback, learning, optimization, and meta-optimization) and whose instances can be composed in progressively more connected ways (single, pipeline, hierarchical, mesh, swarm). Because both dimensions are continuous and, the authors contend, independent, the distance between a conventional DAG workflow and an intelligent swarm of laboratory agents is not an all-or-nothing revolution but a ladder of incremental transitions. The paper's concrete output is a 5×5 evolution matrix that class

Load-bearing premise

The claim that the intelligence and composition dimensions are independent and totally ordered — that raising one level never prevents or requires a particular level of the other — is assumed rather than demonstrated; if that orthogonality fails, the prescribed evolutionary path is not well-defined.

Editorial extensions

If this is right

  • Existing workflow management systems can be extended incrementally: adding observation/feedback turns a static DAG adaptive, adding history turns it learning, and so on, without replacing the execution core.
  • A system's position on the grid dictates its next concrete investments: learning needs data infrastructure, optimizing needs objective/evaluation infrastructure, intelligent needs reasoning engines and knowledge bases; composition upgrades add communication and coordination layers.
  • The 5×5 matrix provides a shared classification language: any current or proposed system, from a batch script to an autonomous lab, can be located on the grid and compared.
  • The architectural blueprint specifies which existing layers (orchestration, coordination, data management, abstraction) must change to support each intelligence/composition level, making the roadmap actionable.
  • If the path is followed, a continuous discovery loop (hypothesis, design, synthesis, characterization, simulation, analysis) can run across federated facilities with human-on-the-loop oversight, compressing discovery cycles by 10–100×.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The independence of the two axes is an assumption rather than a proven theorem; if some intelligence levels cannot compose at certain structural levels, the grid overstates the possible combinations. A concrete test would be to check whether a Learning-level system can genuinely participate in a Swarm without first acquiring Optimizing or Intelligent meta-control.
  • The framework suggests a falsifiable research program: build a benchmark suite of workflow/agent systems classified on the 5×5 grid, then measure whether progressing along one axis changes the difficulty or feasibility of moving along the other. If the axes interact, the roadmap needs a third dimension or a revised ordering.
  • The 100× acceleration figure is a projection from self-driving lab reports, not derived from the framework; a useful extension would be to derive acceleration estimates from the intelligence/composition level, for instance by modeling how much human-in-the-loop latency each upgrade removes.
  • The grid can be read as a design-pattern catalogue for autonomous-science middleware: each cell names a combination of transition-function and composition operators, which could be turned directly into interface specifications.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 6 minor

Summary. The paper proposes a conceptual framework to unify traditional scientific workflows and agentic AI systems, arguing that both can be modeled as finite state machines. It defines two evolution dimensions — intelligence (Static, Adaptive, Learning, Optimizing, Intelligent) and composition (Single, Pipeline, Hierarchical, Mesh, Swarm) — and combines them into a 5×5 evolution matrix that classifies existing systems and charts a roadmap from current DAG-based workflow management systems to fully autonomous, distributed scientific laboratories. The paper also contributes a layered, federated architectural blueprint, discusses challenges, and claims the potential for 100× discovery acceleration. The central thesis is that the path to autonomous science is evolutionary, achieved by incremental upgrades along the two dimensions.

Significance. The state-machine observation is genuinely valuable: it provides a common abstraction that could help the workflows community and the agentic AI community communicate and plan incremental integration. The 5×5 matrix is a memorable organizing device, and the architectural blueprint is concrete enough to guide practical next steps. However, the contribution is essentially a position piece. The formal hierarchy is asserted rather than proven, the key operators Ω and Φ are left without implementation or even precise semantic conditions, and the quantitative acceleration claim is repeated without independent evidence. The strengths are the unifying abstraction and the breadth of examples; the weaknesses are the lack of formal discipline and several internal inconsistencies in the taxonomy. These issues are fixable within the scope of a major revision, and the paper could then serve as a useful roadmap.

major comments (4)
  1. [§3.2, Table 1] The five intelligence levels are not disjoint and are not shown to form a strict chain. 'Learning' updates δ via L(δ,H); 'Optimizing' selects δ* = argmin_δ J(δ). Since any minimizer can be viewed as a learning rule and any learning rule can be framed as optimizing a data-dependent objective, these categories overlap; 'Intelligent' via Ω can express both. The claim that verification complexity goes from 'tractable for static δ' to 'undecidable for metaoptimization Ω' (second paragraph of §3.2) is stated without a model of computation or a proof. At minimum, the paper must give criteria that separate the levels and justify a total order, or explicitly define them as overlapping capabilities and show that the matrix remains well-defined.
  2. [§3.3, Table 3] The taxonomy misinstantiates its own definitions. Swarm is defined as M=Φ({m_i}) with 'local interactions' and 'emergent behavior'; yet Table 3 lists [Static×Swarm] as 'Parameter Sweep'. A parameter sweep has no local interaction and no emergent behavior. Similarly, [Static×Mesh] = 'Fixed Grid' is not an instance of the definition '∀i,j: M_i↔M_j' full connectivity. Each cell must be a genuine instance of the stated composition pattern, or the definitions must be relaxed and the inconsistency acknowledged.
  3. [§3.4] The evolutionary trajectory 'advance intelligence within existing composition, then expand coordination' assumes that Ω and Φ commute and that the two dimensions are independent. The paper does not specify how Ω applies to a composed system M=Φ({m_i}) or whether Ω(Φ({m_i})) = Φ({Ω(m_i)}). Without such a condition, the prescribed path is path-dependent and arbitrary. The paper should either provide a commutation condition, or present the trajectory as one possible strategy rather than a necessary evolutionary sequence.
  4. [Abstract, §6.2] The 100× acceleration figure appears in the abstract and is repeated in §6.2 ('100-fold acceleration') with citation [15], the authors' own AISLE roadmap. No independent evidence or derivation is provided. This strong quantitative claim should be substantiated with demonstrated examples (e.g., the A-lab 50–100× sample throughput cited in §2.3) or explicitly reframed as an aspiration, particularly because it appears in the abstract as a motivating result.
minor comments (6)
  1. [§1] The introduction states that Section 6 discusses 'implications, limitations, and future work', but Section 6 is titled 'Opportunities' and contains no limitations subsection. Correct the pointer or add the missing limitations discussion.
  2. [§3.2] The term 'intelligence' is used both as the name of the dimension and as the label of its highest level ('Intelligent'). This invites confusion; consider renaming the highest level (e.g., 'Meta-optimizing') or using a different dimension name.
  3. [§1] Typo: 'autonomous scientific libraries' should read 'autonomous scientific laboratories'.
  4. [Figure 1 caption] The caption 'the extension of δ with learning L, optimization arg min J, or meta opt. Ω defines its sophistication' is grammatically awkward and should be rephrased.
  5. [Table 1] The notation shifts between the state-machine definition M=(S,Σ,δ,s0,F) and the level definitions: Adaptive and Learning are phrased as properties of δ, while Intelligent is phrased as a transformation of the entire machine M. Clarify how M and δ relate in each row.
  6. [References] Several references are GitHub repositories or blog-style resources (e.g., [11], [24], [32], [48]). This is acceptable for tools, but the formatting should be consistent and, where possible, cite the associated peer-reviewed papers.

Circularity Check

1 steps flagged · score 4.0 of 10

Mild definitional circularity: the claim that autonomous science requires Ω and Φ restates the matrix's own cell labels; the independent state-machine insight and the self-cited 100x estimate are not load-bearing.

  1. self definitional [Section 3.4 (Evolution Matrix), paragraph after Table 3; cf. Table 1 and Table 2 definitions]
    "The bottom-right frontier represents autonomous science. [Intelligent× Swarm] characterizes laboratories where meta-optimization Ω combines with emergence Φ for collective discovery."

    Table 1 defines the Intelligent level as M′=Ω(M,C,G) and Table 2 defines Swarm as M=Φ({m_i}). Table 3 then labels the [Intelligent×Swarm] cell as 'autonomous science.' The paper concludes that 'evolution towards autonomous scientific discovery requires coordinated advancement towards meta-optimization Ω and emergent coordination Φ.' That requirement is not derived from an independent characterization of autonomous science; it is entailed by the paper's own labeling of the target cell. If autonomous science were defined behaviorally (e.g., continuous unattended operation), the necessity of Ω and Φ would be an open question. The state-machine common-denominator observation does not rely on this step, so the circularity is localized to the prescriptive framing.

full rationale

The paper's core observation—that both traditional workflows and AI agents can be modeled as state machines with transition functions and composition patterns—is a modeling claim rather than a derived result, and it does not reduce to the 5×5 matrix. The intelligence and composition dimensions are explicitly introduced as definitions (Section 3.2 and 3.3), so the taxonomy is internally consistent. The only load-bearing step that reduces by construction is the statement that autonomous science requires meta-optimization Ω and emergent coordination Φ, because the [Intelligent×Swarm] cell is defined as 'autonomous science' in Table 3. This is a mild definitional tautology, not an empirical discovery. The 100x acceleration figure is attributed to the authors' own AISLE roadmap [15], but it is a motivational projection and not needed for the framework's validity. No fitted inputs, uniqueness theorems, or ansatz-smuggling self-citations are present. The skeptical concerns about orthogonality and overlap of intelligence levels are correctness risks, not circularity, and do not raise the score further.

Assumptions & free parameters 0 free parameters · 3 assumptions · 3 invented entities

The paper contributes a conceptual synthesis, not an empirical result. It imports several assumptions from prior work, especially the 10-100x acceleration figure from the authors' own AISLE roadmap. The two new formal operators (Ω and Φ) and the two-axis matrix are the paper's own inventions, but they are not anchored to external, falsifiable benchmarks.

assumptions (3)
  • domain assumption All workflows and AI agents can be represented as finite state machines with transition functions δ, L, Ω.
    Presented in Section 3.1 as the 'key insight'; no formal or empirical justification that the representation is faithful, especially for LLM/LRM-based agents.
  • ad hoc to paper Intelligence and composition are independent, cumulative dimensions with five ordered levels each.
    Tables 1 and 2 list levels; the paper assumes a system can advance along one dimension without regressing on the other, which is not argued or tested.
  • domain assumption Autonomous science can accelerate discovery by 10-100x.
    Stated in the abstract and Section 6.2, attributed to reference [15], a roadmap co-authored by the present authors; no independent data is given.
invented entities (3)
  • Intelligence dimension with five levels (Static to Intelligent)
    purpose: Classify and rank workflow autonomy
    Conceptual axis introduced in Section 3.2; no falsifiable prediction.
  • Meta-optimization operator Ω
    purpose: Model self-modifying state machines (Intelligent level)
    Introduced in Section 3.1 as a formal operator; no explicit mechanism or testable consequences.
  • Emergence operator Φ
    purpose: Model swarm-level coordination
    Introduced in Section 3.3; defined informally via 'emergence'; no measurable handle.

how reviews work

0 comments
Cite this review

Pith. "Pith review of The (R)evolution of Scientific Workflows in the Agentic AI Era: Towards Autonomous Science." pith.science (2026). https://pith.science/paper/WZIOTJRE

@misc{pith2026250909915,
  author       = {Pith},
  title        = {Pith review of: The (R)evolution of Scientific Workflows in the Agentic AI Era: Towards Autonomous Science},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WZIOTJRE}},
  note         = {Machine review of arXiv:2509.09915}
}
read the original abstract

Modern scientific discovery increasingly requires coordinating distributed facilities and heterogeneous resources, forcing researchers to act as manual workflow coordinators rather than scientists. Advances in AI leading to AI agents show exciting new opportunities that can accelerate scientific discovery by providing intelligence as a component in the ecosystem. However, it is unclear how this new capability would materialize and integrate in the real world. To address this, we propose a conceptual framework where workflows evolve along two dimensions which are intelligence (from static to intelligent) and composition (from single to swarm) to chart an evolutionary path from current workflow management systems to fully autonomous, distributed scientific laboratories. With these trajectories in mind, we present an architectural blueprint that can help the community take the next steps towards harnessing the opportunities in autonomous science with the potential for 100x discovery acceleration and transformational scientific workflows.

Figures

Figures reproduced from arXiv: 2509.09915 by the authors.

Figure 1
Figure 1. State machine abstraction as a common denomina [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Architectural layers and components: Evolution [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Deployment of architectural components in a fed [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Example of federated scientific discovery: components in the intelligence layer distributed across the infrastructure [PITH_FULL_IMAGE:figures/full_fig_p009_4.png]

Discussion (0). Sign in to comment.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Rethinking Scientific Discovery in the Agentic Era

    cs.CL 2026-07 conditional novelty 5.5 of 10

    SCION claims an agentic OS with Research Execution Plans and layered memory that beats autonomous research-agent baselines on reading, ideation, molecule design, and antibody screening.

  2. Long-range machine-learning potentials with environment-dependent charges enable predicting LO-TO splitting and dielectric constants

    physics.comp-ph 2026-03 unverdicted novelty 5.0 of 10

    Environment-dependent, optionally charge-conserving Coulomb terms grafted onto Moment Tensor Potentials recover LO–TO splitting and experimental dielectric constants for NaCl and improve dimer binding curves.

Reference graph

Works this paper leans on

49 extracted references · 2 linked inside Pith · cited by 2 Pith papers

  1. [15]

    Washburn

    Rafael Ferreira da Silva, Milad Abolhasani, Dionysios A Antonopoulos, Laura Biven, Ryan Coffee, Ian T Foster, Leslie Hamilton, Shantenu Jha, Theresa Mayer, Benjamin Mintz, Robert Moore, Salahudin Nimer, Noah Paulson, Woong Shin, Frédéric Suter, Mitra Taheri, Michela Taufer, and Newell R. Washburn. 2025. A Grassroots Network and Community Roadmap for Inter...

    2025

  2. [1]

    Milad Abolhasani and Eugenia Kumacheva. 2023. The rise of self-driving labs in chemical and materials sciences.Nature Synthesis2, 6 (2023), 483–492

    2023

  3. [2]

    Bryce Allen, John Bresnahan, Lisa Childers, Ian Foster, Gopi Kandaswamy, Raj Kettimuthu, Jack Kordas, Mike Link, Stuart Martin, Karl Pickett, et al . 2012. Software as a service for data scientists.Commun. ACM55, 2 (2012), 81–88

    2012

  4. [3]

    Crusoe, Nebojša Tijanić, Samuel Lampa, et al

    Peter Amstutz, Maxim Mikheev, Michael R. Crusoe, Nebojša Tijanić, Samuel Lampa, et al. 2024. Existing Workflow systems. https://s.apache.org/existing- workflow-systems

    2024

  5. [4]

    Katerina B Antypas, DJ Bard, Johannes P Blaschke, R Shane Canon, Bjoern Enders, Mallikarjun Arjun Shankar, Suhas Somnath, Dale Stansberry, Thomas D Uram, and Sean R Wilkinson. 2021. Enabling discovery data science through cross- facility workflows. In2021 IEEE International Conference on Big Data (Big Data). IEEE, 3671–3680

    2021

  6. [5]

    Katz, Ben Clifford, Rohan Kumar, Lukasz Lacinski, Ryan Chard, Justin M

    Yadu Babuji, Anna Woodard, Zhuozhao Li, Daniel S. Katz, Ben Clifford, Rohan Kumar, Lukasz Lacinski, Ryan Chard, Justin M. Wozniak, Ian Foster, Michael Wilde, and Kyle Chard. 2019. Parsl: Pervasive Parallel Programming in Python. In Proceedings of the 28th International Symposium on High-Performance Parallel and Distributed Computing (HPDC ’19). ACM, 25–36...

    arXiv 2019

  7. [6]

    Rosa M Badia, Laure Berti-Equille, Rafael Ferreira da Silva, and Ulf Leser. 2024. In- tegrating HPC, AI, and Workflows for Scientific Data Analysis (Dagstuhl Seminar 23352).Dagstuhl Reports13, 8 (2024), 129–164

    2024

  8. [7]

    Prasanna Balaprakash, Krishnan Raghavan, Franck Cappello, Ewa Deelman, Anirban Mandal, Hongwei Jin, Imtiaz Mahmud, Komal Thareja, Shixun Wu, Pawel Zuk, et al. 2025. SWARM: Reimagining scientific workflow management systems in a distributed world.The International Journal of High Performance Computing Applications(2025), 10943420251339317

    2025

Show all 49 references
  1. [8]

    Uram, Venkat Vishwanath, and Stefan M

    Prasanna Balaprakash, Michael Salim, Thomas D. Uram, Venkat Vishwanath, and Stefan M. Wild. 2018. DeepHyper: Asynchronous Hyperparameter Search for Deep Neural Networks. In2018 IEEE 25th International Conference on High Performance Computing (HiPC). 42–51. doi:10.1109/HiPC.2018.00014

    2018

  2. [9]

    Andres M Bran, Sam Cox, Oliver Schilter, Carlo Baldassari, Andrew D White, and Philippe Schwaller. 2024. Augmenting large language models with chemistry tools.Nature Machine Intelligence6, 5 (2024), 525–535

    2024

  3. [10]

    Wes Brewer, Ana Gainaru, Frédéric Suter, Feiyi Wang, Murali Emani, and Shantenu Jha. 2025. AI-coupled HPC Workflow Applications, Middleware and Performance. https://arxiv.org/abs/2406.14315. arXiv:2406.14315 [cs.DC]

    2025 arXiv

  4. [11]

    Harrison Chase. 2022. LangChain: Building applications with LLMs through composability. https://github.com/langchain-ai/langchain

    2022

  5. [12]

    Ewa Deelman. 2023. How is artificial intelligence changing science?. In2023 IEEE 19th International Conference on e-Science (e-Science). IEEE, 1–4

    2023

  6. [13]

    Ewa Deelman, Karan Vahi, Mats Rynge, Rajiv Mayani, Rafael Ferreira da Silva, George Papadimitriou, and Miron Livny. 2019. The Evolution of the Pegasus Workflow Management Software.Computing in Science & Engineering21, 4 (2019), 22–36. doi:10.1109/MCSE.2019.2919690

    2019

  7. [14]

    Marco Dorigo, Vittorio Maniezzo, and Alberto Colorni. 1996. Ant system: opti- mization by a colony of cooperating agents.IEEE transactions on systems, man, and cybernetics, part b (cybernetics)26, 1 (1996), 29–41. Conference’17, July 2017, Washington, DC, USA W. Shin, R. Souza...

    1996

  8. [16]

    Badia, Deborah Bard, Ian T

    Rafael Ferreira da Silva, Rosa M. Badia, Deborah Bard, Ian T. Foster, Shantenu Jha, and Frédéric Suter. 2024. Frontiers in Scientific Workflows: Pervasive Integration with HPC.IEEE Computer57, 8 (2024). doi:10.1109/MC.2024.3401542

    2024

  9. [17]

    Ian Foster, Karl Czajkowski, DE Ferguson, Jeffrey Frey, Steve Graham, Tom Maguire, David Snelling, and Steven Tuecke. 2005. Modeling and managing state in distributed systems: The role of OGSI and WSRF.Proc. IEEE93, 3 (2005), 604–612

    2005

  10. [18]

    Holland Hysmith, Elham Foadian, Shakti P Padhy, Sergei V Kalinin, Rob G Moore, Olga S Ovchinnikova, and Mahshid Ahmadi. 2024. The future of self-driving laboratories: from human in the loop interactive AI to gamification.Digital Discovery3, 4 (2024), 621–636

    2024

  11. [19]

    2025.AI Agents in 2025: Expectations vs

    IBM. 2025.AI Agents in 2025: Expectations vs. Reality. IBM. https://www.ibm. com/think/insights/ai-agents-2025-expectations-vs-reality

    2025

  12. [20]

    Anubhav Jain, Shyue Ping Ong, Wei Chen, Bharat Medasani, Xiaohui Qu, Michael Kocher, Miriam Brafman, Guido Petretto, Gian-Marco Rignanese, Geoffroy Hau- tier, Daniel Gunter, and Kristin A. Persson. 2015. FireWorks: a dynamic workflow system designed for high-throughput applica...

    2015 doi

  13. [21]

    James Kennedy and Russell Eberhart. 1995. Particle swarm optimization. In Proceedings of ICNN’95-international conference on neural networks, Vol. 4. IEEE, 1942–1948

    1995

  14. [22]

    Francesc Lordan, Xavier Casas-Moreno, Philip Cummins, Javier Conejero, Rosa M Badia, and Raül Sirvent. 2024. Taming the Swarm: A Role-Based Approach for Autonomous Agents. InEuropean Conference on Parallel Processing. Springer, 15–25

    2024

  15. [23]

    Benjamin P MacLeod, Fraser GL Parlane, Thomas D Morrissey, Florian Häse, Loïc M Roch, Kevan E Dettelbach, Roberto Moreira, Lars PE Yunker, Michael B Rooney, Joseph R Deeth, et al . 2020. Self-driving laboratory for accelerated discovery of thin-film materials.Science Advances6...

    2020

  16. [24]

    Yohei Nakajima. 2023. BabyAGI: An AI-powered task management system. https://github.com/yoheinakajima/babyagi

    2023

  17. [25]

    Azita Nouri, Philip E Davis, Pradeep Subedi, and Manish Parashar. 2021. Exploring the role of machine learning in scientific workflows: Opportunities and challenges. arXiv preprint arXiv:2110.13999(2021)

    2021 arXiv

  18. [26]

    2023.Introducing ChatGPT Enterprise

    OpenAI. 2023.Introducing ChatGPT Enterprise. OpenAI. https://openai.com/ index/introducing-chatgpt-enterprise/

    2023

  19. [27]

    J Gregory Pauloski, Yadu Babuji, Ryan Chard, Mansi Sakarvadia, Kyle Chard, and Ian Foster. 2025. Empowering Scientific Workflows with Federated Agents. arXiv preprint arXiv:2505.05428(2025)

    2025

  20. [28]

    Tabish Rashid, Mikayel Samvelyan, Christian Schroeder, Gregory Farquhar, Jakob Foerster, and Shimon Whiteson. 2018. QMIX: Monotonic value function factori- sation for deep multi-agent reinforcement learning. InInternational conference on machine learning. PMLR, 4295–4304

    2018

  21. [29]

    Jorge Ribeiro, Rui Lima, Tiago Eckhardt, and Sara Paiva. 2021. Robotic process automation and artificial intelligence in industry 4.0–a literature review.Procedia Computer Science181 (2021), 51–58

    2021

  22. [30]

    Loïc M Roch, Florian Häse, Christoph Kreisbeck, Teresa Tamayo-Mendoza, Lars PE Yunker, Jason E Hein, and Alán Aspuru-Guzik. 2020. ChemOS: An orchestration software to democratize autonomous discovery.PLoS ONE15, 4 (2020), e0229862

    2020

  23. [31]

    Russell and Peter Norvig

    Stuart J. Russell and Peter Norvig. 2021.Artificial Intelligence: A Modern Approach (4th ed.). Pearson, Hoboken, NJ. 1166 pages. Global Edition ISBN: 978-1292401133

    2021

  24. [32]

    Significant Gravitas. 2023. AutoGPT: An autonomous GPT-4 experiment. https: //github.com/Significant-Gravitas/AutoGPT

    2023

  25. [33]

    Tyler J Skluzacek, Renan Souza, Mark Coletti, Frederic Suter, and Rafael Fer- reira da Silva. 2024. Towards cross-facility workflows orchestration through distributed automation. InPractice and Experience in Advanced Research Comput- ing 2024: Human Powered Computing. 1–4

    2024

  26. [34]

    Skluzacek, Alexandru Costan, Frédéric Suter, Marta Mattoso, and Rafael Ferreira Da Silva

    Renan Souza, Silvina Caino-Lores, Mark Coletti, Tyler J. Skluzacek, Alexandru Costan, Frédéric Suter, Marta Mattoso, and Rafael Ferreira Da Silva. 2024. Work- flow Provenance in the Computing Continuum for Responsible, Trustworthy, and Energy-Efficient AI. InInternational Conf...

    2024

  27. [35]

    Renan Souza, Amal Gueroudji, Stephen DeWitt, Daniel Rosendo, Tirthankar Ghosal, Robert Ross, Prasanna Balaprakash, and Rafael Ferreira da Silva. 2025. PROV-AGENT: Unified Provenance for TrackingAI Agent Interactions in Agentic Workflows. In5th Workshop on Reproducible Workflow...

    2025

  28. [36]

    OASIS Standard. 2012. Oasis advanced message queuing protocol (amqp) version 1.0.International Journal of Aerospace Engineering Hindawi www. hindawi. com 2018 (2012)

    2012

  29. [37]

    Badia, Bartosz Balis, Kyle Chard, Iacopo Colonnelli, Ewa Deelman, Paolo Di Tommaso, Thomas Fahringer, Carole Goble, Shantenu Jha, Daniel S

    Frédéric Suter, Tainã Coleman, Ílkay Altintas, Rosa M. Badia, Bartosz Balis, Kyle Chard, Iacopo Colonnelli, Ewa Deelman, Paolo Di Tommaso, Thomas Fahringer, Carole Goble, Shantenu Jha, Daniel S. Katz, Johannes Köster, Ulf Leser, Kshitij Mehta, Hilary Oliver, J.-Luc Peterson, G...

    2025

  30. [38]

    Frédéric Suter, Rafael Ferreira da Silva, Ana Gainaru, and Scott Klasky. 2023. Driving Next-Generation Workflows from the Data Plane. In19th IEEE Conference on eScience. doi:10.1109/e-Science58273.2023.10254849

    2023

  31. [39]

    Nathan J Szymanski, Bernardus Rendy, Yuxing Fei, Rishi E Kumar, Tanjin He, David Milsted, Matthew J McDermott, Max Gallant, Ekin Dogus Cubuk, Amil Merchant, et al. 2023. An autonomous laboratory for the accelerated synthesis of novel materials.Nature624, 7990 (2023), 86–91

    2023

  32. [40]

    Domenico Talia. 2002. The open grid services architecture: Where the grid meets the web.IEEE Internet Computing6, 6 (2002), 67–71

    2002

  33. [41]

    2025.ChatGPT: Everything you need to know about the AI chat- bot

    TechCrunch. 2025.ChatGPT: Everything you need to know about the AI chat- bot. https://techcrunch.com/2025/07/18/chatgpt-everything-to-know-about- the-ai-chatbot/

    2025

  34. [42]

    Steven Tuecke, Rachana Ananthakrishnan, Kyle Chard, Mattias Lidman, Brendan McCollam, Stephen Rosen, and Ian Foster. 2016. Globus Auth: A research identity and access management platform. In2016 IEEE 12th International Conference on e-Science (e-Science). IEEE, 203–212

    2016

  35. [43]

    Nicholas Tyler, Robert Knop, Deborah Bard, and Peter Nugent. 2022. Cross-facility workflows: Case studies with active experiments. In2022 IEEE/ACM Workshop on Workflows in Support of Large-Scale Science (WORKS). IEEE, 68–75

    2022

  36. [44]

    Stefanie Warnat-Herresthal, Hartmut Schultze, Krishnaprasad Lingadahalli Shastry, Sathyanarayanan Manamohan, Saikat Mukherjee, Vishesh Garg, Ravi Sarveswara, Kristian Händler, Peter Pickkers, N Ahmad Aziz, et al. 2021. Swarm Learning for decentralized and confidential clinical...

    2021

  37. [45]

    Mark D Wilkinson, Michel Dumontier, IJsbrand Jan Aalbersberg, Gabrielle Apple- ton, Myles Axton, Arie Baak, Niklas Blomberg, Jan-Willem Boiten, Luiz Bonino da Silva Santos, Philip E Bourne, et al . 2016. The FAIR Guiding Principles for scientific data management and stewardshi...

    2016

  38. [46]

    Sean R Wilkinson, Meznah Aloqalaa, Khalid Belhajjame, Michael R Crusoe, Bruno de Paula Kinoshita, Luiz Gadelha, Daniel Garijo, Ove Johan Ragnar Gustafsson, Nick Juty, Sehrish Kanwal, et al. 2025. Applying the FAIR principles to computa- tional workflows.Scientific Data12, 1 (2...

    2025

  39. [47]

    Justin M Wozniak, Rajeev Jain, Prasanna Balaprakash, Jonathan Ozik, Nicholson T Collier, John Bauer, Fangfang Xia, Thomas Brettin, Rick Stevens, Jamaludin Mohd- Yusof, et al. 2018. CANDLE/Supervisor: a workflow framework for machine learning applied to cancer research.BMC Bioi...

    2018

  40. [48]

    Qingyun Wu, Gagan Bansal, Jieyu Zhang, Yiran Wu, Beibin Li, Erkang Zhu, Li Jiang, Xiaoyun Zhang, Shaokun Zhang, Jiale Liu, et al. 2023. AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversation. https://github.com/ microsoft/autogen

    2023

  41. [49]

    Kaiqing Zhang, Zhuoran Yang, and Tamer Başar. 2021. Multi-agent reinforce- ment learning: A selective overview of theories and algorithms.Handbook of reinforcement learning and control(2021), 321–384

    2021

Pith tools

Reviewed August 4, 2026 · model on record in the stance chip above.