{"resourceId":"ascend-personal-hpc-agents-20260926","versions":[{"version":"external-bee6995e436cecc00c11859ae6b27d41649bb088478a73ca0cfffc95ec22b9af","resource":{"id":"ascend-personal-hpc-agents-20260926","title":"ASCEND: laptop-local HPC AI agent matches weather eval within 2.1–2.4%; parallelizes 12,693-line solver ~12h→~2h under bit-for-bit identity","organization":"Authors J. Paul Liu, Uthpala Herath, Andrew Petersen — arXiv preprint (cs.DC)","sector":"Higher education research computing / AI agents for scientific HPC workflows","geography":"United States (campus Slurm/GPU workstation setting in reported cases); transferable methods with site-policy and supervision limits","publishedAt":"September 26, 2026 (arXiv; revised September 30, 2026 v3)","publicationDate":"2026-09-26","eventDate":null,"sourceName":"arXiv 2609.32868","sourceLabel":"Academic preprint (arXiv)","sourceUrl":"https://arxiv.org/abs/2609.32868","evidenceClass":"academic-research","outcomeClass":"mixed","topics":["developers-agents","infrastructure","operating-model","governance-procurement"],"finding":"ASCEND places the agent on the researcher's laptop, reaching Slurm-managed clusters and a GPU workstation over a multiplexed authenticated connection; site-specific execution policies are checked by locally executed tools while the remote language-model agent (Claude Code or Codex) holds no credentials. Under author supervision the authors report four recorded cases: failure-recovery on a planted tensor-device fault; reproduction of a weather-forecasting model evaluation matching published curves to 2.1% (z500) and 2.4% (t850) while recovering an incompletely stated protocol and identifying discrepancies; parallelization of a released 12,693-line geophysical solver under bit-for-bit identity reducing runtime from about twelve hours to about two; and discovery of two undefined-behaviour defects repaired upstream. Separately, a pre-specified policy-validator evaluation rejected 29 of 30 constructed violations and held one for approval while denying 3 of 14 legitimate requests. A controlled end-to-end recovery benchmark remains outstanding. Paper revised 30 Sep 2026 (v3).","sledRelevance":"Campus research-computing programs piloting AI agents for job submission need evidence that credentials and site policy can stay off the remote model—and that bit-for-bit and artifact-preserving loops, not chat fluency, are the acceptance tests.","evidence":"arXiv abs 2609.32868 opened via HTTPS (curl UA SLED-research-bot/1.0); PDF confirmed key metrics. Submission 26 Sep 2026; revised 30 Sep 2026 v3. Findings and excerpt taken from opened abstract/PDF text.","architectureImplications":"Prefer edge/local agent control planes that retain credentials and run site-policy validators locally, with remote LMs proposing but not holding secrets; design Slurm/GPU integrations around multiplexed authenticated sessions and job-level artifact retention.","governanceImplications":"Treat agent autonomy as supervised with explicit policy gates; publish false-positive rates (legitimate denials) alongside violation catch rates before declaring agents production-ready for campus HPC.","securityPrivacyImplications":"Keeping credentials off the remote model reduces cloud-credential exposure, but local tools and multiplexed sessions still require hardened endpoints, audit logs, and least-privilege HPC accounts.","caveats":"Preprint; supervised autonomy; outstanding end-to-end recovery benchmark; policy validator separate from the four cases; single-author-team deployment context.","streamIds":["research"],"roles":{"sales":"Sales framing: Research-computing buyers piloting AI agents for Slurm/GPU work need credential-safe orchestration with measured scientific loops—not another chatbot that holds cluster secrets. ASCEND supports a bounded pitch that a laptop-local agent can recover failed jobs, reproduce a published weather evaluation within about 2.1–2.4%, and speed a large solver under bit-for-bit identity while the remote model holds no credentials. Do not sell unsupervised production autonomy or zero false denials—the policy layer denied 3 of 14 legitimate requests and an end-to-end recovery benchmark remains outstanding.","engineering":"Engineering framing: Fit is a personal agent control plane over Slurm and GPU workstations where credentials and site policies stay local. Integrate multiplexed authenticated transports, locally executed policy validators, job-artifact retention, and pinned scientific environments before enabling multi-step agent loops. Prerequisites include per-resource accounts, policy rulesets, and supervised runbooks. Proof-of-value should replaying bit-for-bit or published-curve checks plus measured policy true/false positives—not demo transcripts alone.","delivery":"Delivery framing: Pilot ASCEND-class agents on one Slurm partition with supervised operators, artifact stores, and a written escalation path when the policy validator blocks legitimate work. Ownership: research computing owns site policies and accounts; PIs own scientific acceptance; security owns credential boundaries. Acceptance requires documented recovery/reproduction cases and policy confusion-matrix review. Risks include supervised-only results being oversold, false denials stalling researchers, and skipping bit-for-bit gates."},"roleDetails":{"sales":{"customerProblem":"Researchers waste time translating intent into Slurm jobs while AI agents that hold cluster credentials create unacceptable risk.","stakeholders":["Research computing / HPC directors","VPR / computational science leads","Security / IAM","PI computational groups","Research software engineers"],"discoveryQuestions":["Where do agent credentials live today?","Which site policies must be enforced before job submit?","What bit-for-bit or published-curve checks define success?","What false-deny rate is acceptable for policy gates?"],"valueHypothesis":"Local-control HPC agents with measured recovery/reproduction loops can cut scientific workflow toil without exporting cluster credentials to remote models.","potentialEngagement":"Four-week supervised pilot on one Slurm partition with recovery, reproduction, and policy confusion-matrix report.","unsupportedClaims":["Unsupervised production autonomy","Zero false denials from policy gates","Guaranteed multi-site generalization","Replacement for research software engineers"]},"engineering":{"fit":"Laptop-local agent over Slurm/GPU with remote LM proposing actions and local tools enforcing policy/credentials.","architectureIntegration":"Wire multiplexed authenticated sessions, local policy validators, job artifact stores, and pinned runtime environments; keep secrets off the remote LM.","prerequisites":["Per-resource HPC accounts","Site policy ruleset","Artifact/logging store","Supervised operator runbook"],"constraints":["Author-supervised evaluation","Outstanding end-to-end recovery benchmark","Policy false denials (3/14 legitimate)","Preprint evidence"],"security":"Never place cluster credentials in the remote model; audit local tool actions and enforce least-privilege accounts.","proofOfValue":["Published-curve or bit-for-bit match on a target workload","Documented failure-recovery loop with artifacts","Policy validator TPR/FPR on constructed cases","Credential boundary review"]},"delivery":{"work":"Stand up supervised pilot, encode site policies, retain job artifacts, run reproduction/recovery drills, review policy false denials.","dependencies":["Slurm/GPU access","Security review","PI scientific validators","Logging/storage"],"ownership":"Research computing owns policies/accounts; security owns credential boundaries; PIs own scientific acceptance.","skillsAdoption":"Train operators and RSEs on agent supervision, policy exceptions, and bit-for-bit acceptance checks.","governanceCheckpoints":["Security architecture review","Pilot scope freeze","Mid-pilot policy false-deny review","Go/no-go for wider rollout"],"acceptanceCriteria":["≥1 documented recovery loop","≥1 reproduction/bit-for-bit report","Policy confusion matrix reviewed","No remote holding of cluster credentials"],"risks":["Overselling supervised results","Policy false denials","Artifact gaps","Scope creep to unsupervised autonomy"]}},"retrievedAt":"2026-09-30T22:02:14-05:00","enrichedAt":"2026-09-30T22:08:00-05:00","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Personal agents can lower the Slurm expertise barrier for domain scientists, but still require trained supervisors and clear exception paths when policy gates block legitimate work.","procurementImplications":"When buying AI-for-HPC agent tooling, require local credential custody, site-policy hooks, artifact retention, and measured reproduction/recovery tests—not demo-only fluency.","operatingModelImplications":"Operating model is supervised agent ops with policy confusion-matrix monitoring, not unattended campus-wide agent submission.","sourceVerification":{"openedUrl":"https://arxiv.org/abs/2609.32868","referenceExcerpt":"matching the published curves to 2.1 percent on z500 and 2.4 percent on t850","promptVersion":"sled-research-v4.0","model":null,"basis":"agent-reported inspection"}}}]}