Thinking Fast, Anywhere: One System-1 for the Decision Points of Any Agent Workflow
Organizations: University of Science and Technology of China · Nanyang Technological University
Abstract
Autonomous LLM agents incur substantial cost and latency by invoking a large model (System-2) at every step, even when many intermediate actions choose among a small set of known candidates. Existing dual-system agents offload such decisions to a lightweight System-1 model, but rely on workflow-specific hooks authored per environment. We present 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. 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. 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. 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.
Figures & tables
| Public skills | 1,072 |
|---|---|
| Skill categories | 16 |
| Decision points (D / E / G, %) | 17,865 (21.3 / 64.5 / 14.2) |
| Skills with 1 E node | 100% |
| Synthesized tasks | 2,846 |
| Consensus states | 25,986 |
| Teacher agreement | 89.2% |
| SOPBench | ALFWorld | WebShop | Avg. | |||||||||
| Configuration | Pass | Sav. % | Time (s) | SR | Sav. % | Time (s) | Score | SR | Sav. % | Time (s) | pt | Sav. % |
| System-2: DeepSeek-V4.1-Flash | ||||||||||||
| System-2 agent with skills | 55.2 | – | 51.4 | 82.8 | – | 163 | 54.9 | 34.4 | – | 146 | 17.8 | – |
| Compiled workflow (All-S2) | 83.2 | 86 | 25.0 | 96.3 | 81 | 21.3 | 66.9 | 38.0 | 76 | 29.0 | 0.0 | 81 |
| Jev as System-1 † | 84.0 | 87 | 23.9 | 96.3 | 84 | 20.5 | 67.1 | 38.6 | 83 | 17.3 | +0.3 | 84 |
| AnyS1 warm | 84.3 | 91 | 15.6 | 95.5 | 88 | 12.9 | 68.1 | 38.8 | 86 | 17.1 | +0.5 | 89 |
| (a) Runtime deferral gate | SOP (Cost) | ALF (Cost) | WS (Cost) |
|---|---|---|---|
| AnyS1 warm | 84.3 (0.09 ) | 95.5 (0.12 ) | 68.1 (0.14 ) |
| AnyS1 cold | 83.2 (0.13 ) | 95.5 (0.15 ) | 67.9 (0.11 ) |
| All-S1 | 84.3 (0.09 ) | 40.3 (0.04 ) | 66.3 (0.02 ) |
| (b) Training state source | Dev | SOP E | ALF E |
| Any-DP | 83.8 | 93.5 | 68.9 |
| LLM-written synthetic | 65.0 | 91.7 | 63.6 |
Appendix figures & tables6 assets
Supplementary material from the paper’s appendix.
Appendix
| Dev | SOP | ALFWorld | ||
| Model | E | E | search | |
| Zero-shot baselines | ||||
| Kev-4B | 57.8 | 88.7 | 56.9 | 16.3 |
| AgentDecision-4B | 74.6 | 88.2 | 69.5 | 40.4 |
| Qwen3.5-4B | 71.4 | 91.7 | 70.9 | 44.0 |
| Qwen3.5-27B | 85.9 | 93.8 | 77.2 | 56.8 |
| AURC | AnyS1 warm : agree / errors per 100 | ||||
|---|---|---|---|---|---|
| Benchmark | Gate | Random | Delegated | Gate | Random |
| SOPBench | 0.014 | 0.086 | 98.0% | 92.7 / 7.1 | 91.4 / 8.4 |
| ALFWorld | 0.100 | 0.318 | 50.4% | 92.4 / 3.8 | 68.2 / 16.0 |
| WebShop | 0.156 | 0.302 | 41.4% | 84.3 / 6.5 | 69.8 / 12.5 |
LLM Agent Workflow Optimization
LLM agent workflow optimization studies how to design and run multi-step agents for better task success, reliability, latency, and cost.
See all 286 papers