Li, Rui, He, Liyang, Zhang, Zheng, Huang, Zhenya, Liu, Qi · Nov 18, 2026
Thinking Fast, Anywhere: One System-1 for the Decision Points of Any Agent Workflow
Markdown · JSON · Original source
Rui Li, Liyang He, Zheng Zhang, Zhenya Huang and Qi Liu present AnyS1, a framework that turns procedural skills—written instructions for carrying out tasks—into decision points a shared model can handle. Their premise is that an agent need not invoke a large System-2 model at every step when a choice is among known options. AnyS1 uses a lightweight System-1 model for routine choices, then falls back to System-2 when confidence is too low. The authors say they built the system from more than 1,000 public skills and trained a 4B pointer System-1 model. Across three unseen and different System-2 model families, they report that AnyS1 preserved task reliability while cutting episode cost by 75–91% and model-call by 63–92% relative to an uncompiled System-2 agent. They also report performance under threshold transfer. The claimed benefit is fewer costly, slow calls to a large model without sacrificing task reliability in those evaluations. The abstract does not name the benchmarks or show how reliability and costs were measured, so the reported results cannot be assessed in detail from this evidence alone.
Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.
Original source
cs.AI · Article
Source details
Original source date: Nov 18, 2026 (2026-11-18).
Date basis: Date basis not provided.
Added to Escape Meridian: 2026-11-19T08:34:13.728Z. First observed: 2026-11-19T05:04:12.387Z. Retrieved: 2026-11-19T08:32:58.010Z.
Referenced by
Recorded citations in public readings. A reference alone does not establish agreement or influence.
No public references recorded here yet.
Limits
The supplied evidence is the abstract; it does not name the three or describe how task reliability and costs were measured. The reported results are the authors’ claims about evaluations; this evidence does not independently establish performance in other settings.
Open question
Which three unseen were used, and how did the authors measure task reliability and episode cost?
Source-stated
The authors present AnyS1 as a framework that converts procedural skills into a shared source of decision points.
AnyS1, a framework that turns human-authored procedural skills, whose instructions naturally specify where choices arise and which options apply, into a shared, domain-general source of decision points.
Source-stated
The authors say AnyS1 uses a lightweight System-1 model for routine choices and falls back to System-2 when confidence is below a calibrated threshold.
At run time, compiled workflows resolve routine decisions in one forward pass and fall back to System-2 when confidence falls below a calibrated threshold.
Source-stated
The authors report building the system using more than 1,000 public skills and training a 4B pointer System-1 model.
Offline, AnyS1 mines grounded enumerable decisions across over 1,000 public skills, gathers execution-grounded trajectory states inside isolated containers to construct Any-DP, and trains a single calibrated 4B pointer System-1 with label smoothing.
Source-stated
Across three unseen benchmarks and different System-2 model families, the authors report preserved task reliability, lower episode cost and model-call time, and performance under zero-shot threshold transfer.
Across three unseen benchmarks and distinct System-2 model families, AnyS1 preserves task reliability while cutting episode cost by 75-91% and model-call time by 63-92% relative to an uncompiled System-2 agent, and maintains performance under zero-shot threshold transfer.
Source and contributor context
Li, Rui, He, Liyang, Zhang, Zheng, Huang, Zhenya, Liu, Qi · Article author
These are credits stated by the source; they do not establish a profile match or authorship of every passage.
Source media
No suitable media was found by the supported page extraction.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
Benchmark
A defined set of tests used to compare systems. Results depend on the tasks and measures selected.
Zero-shot
Doing a task without worked examples (zero-shot) or with only a few examples given in the prompt (few-shot).
arXiv
arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
TIME
TIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services.
Editorial provenance
Version 1. Publication ID: 3c6748b2-e3df-4a4b-85a9-02bf24f29f1a.
Prepared by gpt-6-luna; grounding review by gpt-6-luna.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-19T08:34:13.728Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: material_development.
Observation: 951d623c0fbe9ab9824d110d054a564514cf3a44babf6ab85226e14c05085d4d. Source text hash: 66da8f0bb64a45f4bc2085592d39a34a16f784c1297c783e54aa377b61b47c17.
Revision and source-check history
Version 1 · 2026-11-19T08:34:13.728Z
Thinking Fast, Anywhere: One System-1 for the Decision Points of Any Agent Workflow
Rui Li, Liyang He, Zheng Zhang, Zhenya Huang and Qi Liu present AnyS1, a framework that turns procedural skills—written instructions for carrying out tasks—into decision points a shared model can handle. Their premise is that an agent need not invoke a large System-2 model at every step when a choice is among known options. AnyS1 uses a lightweight System-1 model for routine choices, then falls back to System-2 when confidence is too low. The authors say they built the system from more than 1,000 public skills and trained a 4B pointer System-1 model. Across three unseen and different System-2 model families, they report that AnyS1 preserved task reliability while cutting episode cost by 75–91% and model-call by 63–92% relative to an uncompiled System-2 agent. They also report performance under threshold transfer. The claimed benefit is fewer costly, slow calls to a large model without sacrificing task reliability in those evaluations. The abstract does not name the benchmarks or show how reliability and costs were measured, so the reported results cannot be assessed in detail from this evidence alone.
Original source
cs.AI · Article
Source details
Original source date: Nov 18, 2026 (2026-11-18).
Date basis: Date basis not provided.
Added to Escape Meridian: 2026-11-19T08:34:13.728Z. First observed: 2026-11-19T05:04:12.387Z. Retrieved: 2026-11-19T08:32:58.010Z.
Limits
The supplied evidence is the abstract; it does not name the three or describe how task reliability and costs were measured. The reported results are the authors’ claims about evaluations; this evidence does not independently establish performance in other settings.
Open question
Which three unseen were used, and how did the authors measure task reliability and episode cost?
Source-stated
The authors present AnyS1 as a framework that converts procedural skills into a shared source of decision points.
AnyS1, a framework that turns human-authored procedural skills, whose instructions naturally specify where choices arise and which options apply, into a shared, domain-general source of decision points.
Source-stated
The authors say AnyS1 uses a lightweight System-1 model for routine choices and falls back to System-2 when confidence is below a calibrated threshold.
At run time, compiled workflows resolve routine decisions in one forward pass and fall back to System-2 when confidence falls below a calibrated threshold.
Source-stated
The authors report building the system using more than 1,000 public skills and training a 4B pointer System-1 model.
Offline, AnyS1 mines grounded enumerable decisions across over 1,000 public skills, gathers execution-grounded trajectory states inside isolated containers to construct Any-DP, and trains a single calibrated 4B pointer System-1 with label smoothing.
Source-stated
Across three unseen benchmarks and different System-2 model families, the authors report preserved task reliability, lower episode cost and model-call time, and performance under zero-shot threshold transfer.
Across three unseen benchmarks and distinct System-2 model families, AnyS1 preserves task reliability while cutting episode cost by 75-91% and model-call time by 63-92% relative to an uncompiled System-2 agent, and maintains performance under zero-shot threshold transfer.
Source and contributor context
Li, Rui, He, Liyang, Zhang, Zheng, Huang, Zhenya, Liu, Qi · Article author
These are credits stated by the source; they do not establish a profile match or authorship of every passage.
Source media
No suitable media was found by the supported page extraction.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
Benchmark
A defined set of tests used to compare systems. Results depend on the tasks and measures selected.
Zero-shot
Doing a task without worked examples (zero-shot) or with only a few examples given in the prompt (few-shot).
arXiv
arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
TIME
TIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services.
Editorial provenance
Version 1. Publication ID: 3c6748b2-e3df-4a4b-85a9-02bf24f29f1a.
Prepared by gpt-6-luna; grounding review by gpt-6-luna.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-19T08:34:13.728Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: material_development.
Observation: 951d623c0fbe9ab9824d110d054a564514cf3a44babf6ab85226e14c05085d4d. Source text hash: 66da8f0bb64a45f4bc2085592d39a34a16f784c1297c783e54aa377b61b47c17.
uncertain · 2026-11-19T08:34:13.728Z · anthropic/claude-opus-5.5
Publication: 3c6748b2-e3df-4a4b-85a9-02bf24f29f1a
attribution — pass: The reading correctly names the five listed authors and attributes the framework and results to them as the authors' own reported claims.
Authors:Rui Li, Liyang He, Zheng Zhang, Zhenya Huang, Qi Liu
summary — uncertain: The summary omits the paper's central contrast. Existing dual-system agents already hand off simple choices to System-1, but they need workflow-specific hooks written for each environment; AnyS1 aims to make this domain-general. Without that, the 'Anywhere' idea is lost. Smallest repair: add one sentence on the per-environment hooks. 'They also report performance under zero-shot threshold transfer' drops the claim that performance is maintained; say 'maintains performance'. Central terms such as 4B pointer, uncompiled System-2 agent, episode cost and zero-shot threshold transfer are not explained in plain language; add tooltips or brief glosses.
Existing dual-system agents offload such decisions to a lightweight System-1 model, but rely on workflow-specific hooks authored per environment.
date — pass: originalPublished 2026-11-18 matches the arXiv v1 submission date. The feed date of Nov 19 is a later announcement.
[Submitted on 18 Nov 2026]
support — pass: The claims, the reported percentages and the stated benefit, limited to those evaluations, are grounded in the abstract. The limitation and question about unnamed benchmarks and measurement methods correctly reflect gaps in the evidence. Nothing implies real-world reliability beyond the reported tests.
AnyS1 preserves task reliability while cutting episode cost by 75-91% and model-call time by 63-92% relative to an uncompiled System-2 agent
media — pass: The selected media is the verified direct arXiv abstract page for this paper, which is the most direct access to the source.
Challenge or correct this reading
Send a correction or an opposing source with this reading’s URL. Disagreement and uncertainty can remain unresolved.
A defined set of tests used to compare systems. Results depend on the tasks and measures selected.
Definition referenceTIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services.
Explore TIMEDoing a task without worked examples (zero-shot) or with only a few examples given in the prompt (few-shot).
Definition referencearXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXivA defined set of tests used to compare systems. Results depend on the tasks and measures selected.
Definition referenceA defined set of tests used to compare systems. Results depend on the tasks and measures selected.
Definition referenceA defined set of tests used to compare systems. Results depend on the tasks and measures selected.
Definition referenceTIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services.
Explore TIMEDoing a task without worked examples (zero-shot) or with only a few examples given in the prompt (few-shot).
Definition referencearXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXivA defined set of tests used to compare systems. Results depend on the tasks and measures selected.
Definition referenceA defined set of tests used to compare systems. Results depend on the tasks and measures selected.
Definition reference