Agent workflows compiled into consistent reusable neuro symbolic programs
FlashScript: Compiling Recurring Agent Workflows into Reusable Neuro-Symbolic Programs
Human-Computer InteractionArtificial IntelligenceSoftware Engineering
Summary
When computer agents do repetitive tasks, they often run their steps anew each time, sometimes making different mistakes or wasting time. These authors created a way to turn a successful run into a reusable program that mixes coded steps with smart calls to language models. This makes the process faster, consistent, and easier to check because only the parts needing judgment rely on AI, while the rest runs as clear code. Their approach worked well on many test tasks, producing stable and correct results much faster than before.
What this means in practice
- •For workflow automation engineers: Create consistent and faster automations by compiling repeated agent tasks into reusable neuro-symbolic programs that reduce AI calls and runtime variability.
- •For business intelligence teams: Use compiled neuro-symbolic programs to reliably rerun analytics workflows across updated data without inconsistencies or recalculating all steps.
Authors
Sadra Sabouri, Hosein Hasanbeig, Adam Sage, Shreya Sinha, Cale Darling, Ashish Tiwari, Katie Garms, Souti Chattopadhyay, Sumit Gulwani
Abstract
Many agent tasks recur with new data or small changes, yet agents regenerate their reasoning and actions on each run, producing different results. Distilling a prior run into a skill helps the agent avoid earlier mistakes, but it still re-runs the full process each time, so runs can vary and incur that cost. We compile a successful agent run into a reusable neuro-symbolic program (NSP) with a set of guidelines we call FlashScript. Actions and data transformations become code, instance-specific values become parameters, and decisions that need model judgment become LLM calls. We identify three layers where these calls belong and three techniques that replace most of them with code. The compilation does not remove nondeterminism; it localizes it, so the deterministic program and its neural calls can be reviewed and tested separately. Across 120 WorkBench analytics tasks with Claude Haiku 4.5, the NSPs were correct and made identical tool calls on every run, using about 70 times fewer input tokens and a third of the runtime of a skill-guided agent whose runs were inconsistent. In a composite version of the tasks, the skill-guided agent's accuracy fell to 23%, and its output inconsistency increased, while the NSPs remained correct and identical across all runs. The NSPs successfully generalized to all 108 unseen instances of the same templates. Our findings suggest that NSPs from agent trajectories can be considered cumulative software assets for end-users to test, maintain, and reuse in their recurring workflows.