Report of the 2026 Workshop on Next-Generation Ecosystems for Scientific Computing: Harnessing Community, Software, and AI for Cross-Disciplinary Team Science

Online access

• DOE reports produced after 1991 and a growing number of pre-1991 documents are available free at OSTI.GOV.OSTI.GOV is a service of the U.S. Department of Energy’s Office of Scientific and Technical Information.Source: DOCUMENT AVAILABILITY · p. 2

Nondigital reports

• Nondigital reports may be purchased by the public from NTIS or obtained by DOE and DOE contractors from OSTI.The passage specifies different access channels for the public and for DOE and its contractors.Source: DOCUMENT AVAILABILITY · p. 2

Disclaimer

• The report disclaims warranties and liability, says product references do not imply government endorsement, and notes authors’ views need not reflect listed institutions.The disclaimer names the U.S. Government, its agencies, UChicago Argonne, LLC, and Argonne National Laboratory.Source: DOCUMENT AVAILABILITY · p. 2

Executive Summary: Toward Strategies for Trustworthy, AI-Enabled Scientific Computing

• The 2026 workshop was the second in a three-year series and shifted from identifying challenges toward examining actions to address them.Advances in AI, heterogeneous computing, emerging quantum capabilities, automation, and data-intensive research are reshaping scientific tools, institutions, workforce models, and collaborative practices.Source: Executive Summary: Toward Strategies… · p. 6
• The workshop identified four interdependent themes: AI-enabled discovery software ecosystems; trust, validation, and traceability; human-AI teaming and paradigm shifts; and workforce, pedagogy, and governance.Source: Executive Summary: Toward Strategies… · p. 6
• Weak validation, insufficient human oversight, or inadequate skills and governance can undermine trust, obscure errors, or make technical advances difficult to sustain.Future environments must enable people to understand, direct, evaluate, and intervene in computational processes as AI takes larger roles in scientific workflows.Source: Executive Summary: Toward Strategies… · p. 6
• The report translates the workshop’s themes into eight community-action priorities for adapting to technological change while preserving scientific rigor and public trust.Together, the priorities aim to support greater speed and efficiency, deeper creativity, stronger collaboration, and more reliable pathways to discovery.Source: Executive Summary: Toward Strategies… · p. 6

Background and Motivation

• The workshop advances a three-year series from identifying challenges to strategic priorities, making trust and governance explicit as agentic AI increases the urgency of persistent ecosystem issues.It refines earlier challenge areas into four interdependent themes and eight priorities for community action.Source: §1 · p. 7
• Scientific computing ecosystems combine technical, social, and organizational structures that must adapt and improve together through expertise, institutional stewardship, and sustained cross-sector contributions.The report includes government, academia, national laboratories, nonprofits, and the private sector among those whose contributions support ecosystem sustainability.Source: §1 · p. 8
• AI is shifting scientific workflows from tool use toward active participation, heightening the need for provenance, validation, attribution, accountability, and records of decisions and computational processes.The shift does not reduce the need for human expertise; it changes where that expertise is needed.Source: §1 · p. 8
• The workshop aims for better science rather than simply faster science, aligning technology, user experience, team norms, workforce development, governance, and incentives around valid, useful, scrutinizable insights.Human judgment remains important for framing questions, integrating expertise, selecting methods, interpreting results, and managing uncertainty.Source: §1 · p. 8
• The report’s themes and priorities are a qualitative synthesis of presentations, panels, breakout discussions, plenary report-outs, and participant review, not a formal consensus process or systematic literature review.The writing team consolidated recurring concerns, proposed actions, and areas for further study through multiple review rounds.Source: §1 · p. 9

Software ecosystems for AI-enabled scientific discovery.

• Shared research assets capture knowledge, context, and operational experience for reuse across projects and communities, serving as infrastructure that helps distributed teams understand, extend, validate, and trust complex scientific computing workflows.Digital-twin and digital-thread frameworks connect models, data, software, and lifecycle context through traceable, reusable representations. Scientific computing ecosystems should help researchers select computational approaches suited to the problem, scientific objectives, validation requirements, and available resources [74] [75].Source: Software ecosystems for AI-enabled… · p. 10
• Usable ecosystems should guide scientists from their problems to relevant resources and next steps, while directing scrutiny toward actions with greater uncertainty, risk, cost, or consequence.User experience affects participation and tool adoption; insufficient guidance can produce plausible but scientifically invalid results, and researchers must be able to trace, evaluate, and challenge system outputs.Source: Software ecosystems for AI-enabled… · p. 10

Trust, validation, and traceability.

• Trust requires workflow-wide VVUQ, provenance, traceability, transparency, and auditability that capture decisions, assumptions, processes, and human contributions.VVUQ must extend to hybrid workflows and assess confidence, limitations, validity regimes, and decision implications.Source: Trust, validation, and traceability · pp. 10–11
• Scientific software can produce misleading results despite executing without errors, so conventional tests need mathematically or statistically grounded diagnostics and, where feasible, formal reasoning.When exact reproducibility is impossible, traceability and uncertainty quantification help judge credibility, while auditability makes these practices reviewable and actionable.Source: Trust, validation, and traceability · p. 11

Human-AI teaming and paradigm shifts.

• Human-AI teaming requires balancing automation with augmentation, increasing transparency, validation, and human judgment as uncertainty, risk, or scientific consequence rises.Agentic workflows also need shared goals and task states, exposed assumptions and limitations, preserved provenance, reconciliation of conflicting outputs, and escalation when expertise or authority is needed [81].Source: Human-AI teaming and paradigm shifts · p. 11
• Human-AI teams need explicit norms for automation, review, and responsibility, and should be evaluated for scientific validity, error detection, reproducibility, recovery, collaboration, and researchers’ understanding.Human researchers remain central to framing questions, integrating knowledge, evaluating evidence, managing uncertainty, and interpreting unexpected results.Source: Human-AI teaming and paradigm shifts · pp. 11–12

Workforce, pedagogy, and governance.

• Workforce preparation must build judgment beyond technical proficiency through multiple career-stage pathways and durable taskwork and teamwork competencies.Near-term training can emphasize durable competencies and responsible tool use; medium-term efforts can build human-AI teaming experience; longer-term work should study changes to roles, institutions, and careers.Source: Workforce, pedagogy, and governance · pp. 12–13

• Build and sustain shared research assets and open infrastructure.

• Treat curated libraries, datasets, standards, pipelines, benchmarks, validation examples, and documentation as shared infrastructure, and value the people who maintain and improve them.Maintaining these assets supports reuse, continuity, and long-term sustainability.Source: • Build and sustain shared research… · p. 13

• Build capabilities for trust and traceability.

• Build VVUQ, provenance, and auditability into scientific processes, tracing assumptions, decisions, human interventions, and execution context while addressing agent identity, authorization, security, and recovery.Traceability supports review, reproduction where possible, and assessment when exact reproduction is not feasible.Source: • Build capabilities for trust and… · p. 13

• Treat user experience as scientific infrastructure.

• Design interfaces, access, support, and policies to help researchers identify methods, understand constraints, interpret outputs, and know when validation is needed while preserving scrutiny for significant risks.Effective user-experience design can lower unnecessary barriers when uncertainty, cost, or consequences are significant.Source: • Treat user experience as scientific… · pp. 13–14

• Establish explicit norms for human-AI teaming.

• Set explicit human-AI teaming norms for delegation, expert escalation, approval of AI-generated results, and shared responsibility, supported by boundary-spanning roles across disciplines and organizations.Source: • Establish explicit norms for… · p. 14

• Center workforce development on judgment.

• Prepare students, practitioners, and institutional leaders in collaboration, verification, reasoning under uncertainty, and responsible tool use; use pilots and longitudinal studies to examine changing roles and careers.Technical training alone is insufficient; studies should also examine team behavior and job satisfaction.Source: • Center workforce development on… · p. 14

• Coordinate across sectors and borders.

• Sustain coordination across academia, laboratories, industry, government, nonprofits, and international partners to align standards, maintain common infrastructure, share lessons, and reduce duplication.Source: • Coordinate across sectors and borders · p. 14

• Evaluate scientific value.

• Evaluate whether AI-enabled ecosystems improve scientific and engineering practice through insight validity, transparency, validation, human intervention, sustained support, collaboration, and breadth of inquiry.Near-term work should prioritize shared assets, trust infrastructure, human-AI norms, and stewardship incentives; pilots and longitudinal studies should test other priorities. The synthesis reflects recurring workshop themes, not formal consensus or a systematic review, and priorities should be refined across contexts.Source: • Evaluate scientific value · p. 14

4 Conclusion

• Scientific computing’s future depends on ecosystems, not isolated tools, and AI’s value lies in supporting valid, interpretable, and creative discovery rather than speed alone.The conclusion says these needs persist as AI systems, computing architectures, and development practices change.Source: §4 · p. 14
• Meeting this standard requires reusable software, auditable workflows, robust VVUQ and traceability, boundary-spanning teams, critical-judgment workforce pathways, and incentives for stewardship.These requirements remain as AI systems, computing architectures, and development practices change.Source: §4 · pp. 14–15
• Socio-technical co-design must evolve people, software, data, infrastructure, institutions, incentives, and norms together through feedback, adaptation, and stewardship.Done well, such ecosystems can accelerate discovery and expand inquiry while preserving rigor, transparency, accountability, and collaboration.Source: §4 · p. 15

A Workshop Description

• Held April 14–16, 2026, in Chicago, the workshop engaged over 45 cross-disciplinary experts to chart a path toward more powerful, sustainable, and collaborative scientific software ecosystems.The description is reproduced from pre-workshop materials and retains its original future-tense wording.Source: A Workshop Description · p. 24

Strategy

• The workshop’s co-design approach weaves team-based software, AI-enabled workflows and infrastructure, and community and workforce development to advance technical innovation and community evolution together.The series builds on the 2025 workshop and aims for broad, sustainable, impactful scientific computing ecosystems.Source: Workshop Charge · p. 24

Areas of emphasis

• New scientific frontiers, AI-augmented workflows, and heterogeneous platforms call for new software and workforce approaches, with community practices reflecting broad expertise, use cases, and stakeholder needs.Source: Workshop Charge · p. 25
• The workshop seeks robust, interoperable, scalable software ecosystems for AI-driven discovery in HPC and data-intensive settings, with tools and infrastructure supporting openness, reproducibility, and broad participation.Source: Workshop Charge · p. 25
• It addresses evolving roles, career paths, and practices for cross-disciplinary software teams, emphasizing ongoing co-design dialogue among scientists, developers, users, and maintainers.Source: Workshop Charge · p. 25
• The workshop explores AI-assisted development, testing, tuning, documentation, and maintenance to improve productivity and sustainability, guided by feedback from real-world workflows and community needs.Source: Workshop Charge · p. 25
• It aims to cultivate teams equipped for AI-rich scientific computing and foster collaborative cultures that bridge technical excellence with community-centered practice.Source: Workshop Charge · p. 25

Workshop Objectives

• The workshop seeks shared understanding of emerging methodologies, tools, and norms for team-based software development, integrating technical rigor with community sustainability.It also aims to curate and refine resources, frameworks, and exemplars that support excellence, effectiveness, and resilience in scientific software collaborations.Source: Workshop Objectives · p. 25
• The workshop aims to identify strategies to address barriers facing scientific software teams while enabling innovation aligned with emerging research and AI-driven needs.It also seeks a shared vision for next-generation ecosystems that harmonizes technical innovation with community dynamics.Source: Workshop Objectives · p. 26
• The workshop seeks concrete actions to strengthen and broaden the workforce and foster durable communities prepared for urgent challenges in HPC and AI-enabled scientific computing.It envisions cross-disciplinary ecosystems driving scientific discovery through high-quality software that is co-designed, AI-enabled, and sustainably maintained.Source: Workshop Objectives · p. 26

• Participants will use discussion and co-design to examine emerging practices in AI-enabled team software, assess progress and gaps, and prioritize technical, organizational, and community focus areas.Source: Workshop Outcomes · p. 26
• The post-workshop report will extend the 2025 workshop’s findings and inform scientific software development, workforce advancement, and ecosystem coordination.It may also help shape evolving perspectives, policies, and practices across the scientific computing community.Source: Workshop Outcomes · p. 26
• Interweaving community considerations with technical discussions aims to strengthen the scientific software community, foster cross-disciplinary collaboration, and accelerate scientific discovery.These aims span the three-year workshop effort in an increasingly AI-driven world.Source: Workshop Outcomes · p. 26

C Workshop Agenda

• The three-day agenda progresses from identifying ecosystem changes and dependencies to developing strategies, prototypes, roadmaps, and measures of success.Day 1 asks groups to identify technical and social changes; later sessions develop proposals with metrics and dependencies and discuss how to evaluate progress.Source: C Workshop Agenda · pp. 29–31
• The workshop’s cross-disciplinary discussions address software and next-generation science, AI-driven ecosystems, team-based research, software productivity and sustainability, and community and workforce development.Source: C Workshop Agenda · p. 29
• Day 2’s open forum emphasizes synergies and co-design opportunities while compiling changes in scientific computing ecosystems and their dependencies.Source: C Workshop Agenda · p. 30
• Day 3 focuses on potential paths forward and asks participants to define success, develop strategies, and evaluate progress across ecosystem topics.The agenda includes a plenary panel on challenges in building the future workforce and communities.Source: C Workshop Agenda · p. 31
• Agentic AI makes existing human-centered norms for reproducibility, peer review, workforce training, attribution, and funding increasingly inadequate for human-agent research teams.The talk calls for auditable software architectures, disclosure norms for agent-generated methods, and workforce pipelines that train scientists to critically oversee AI agents.Source: C Workshop Agenda · p. 32

ARGONNE NATIONAL LABORATORY

• Argonne is a U.S. Department of Energy research facility operated by UChicago Argonne, LLC, and is the Midwest’s largest federally funded R&D facility.It is located in Lemont, Illinois, about 25 miles (40 km) southwest of Chicago, and conducts basic and applied research.Source: ARGONNE NATIONAL LABORATORY · p. 33