Spine improves coding AI by organizing tasks in a structured tree
Spine: Compiling Subtask Trees into Structured Context for Long-Horizon Coding Agents
Software EngineeringArtificial Intelligence
Summary
Long coding projects have many small steps inside them, which can confuse AI helpers when everything is saved as one long story. The authors created Spine, a way for AI to keep track of these steps as a tree instead of a list. This helps the AI focus only on unfinished parts and remember results of finished parts without losing details. Their system made AI coding assistants more accurate and cost-effective in tests. They also made Spine work with popular AI models, improving their performance too.
What this means in practice
- •For software development teams: Enable coding AI tools to manage complex projects by tracking subtasks more efficiently to increase accuracy and reduce run costs.
- •For ai service operators: Integrate structured subtask management to scale coding assistant services with improved performance and reduced operational expenses.
Authors
Jiahong Xiang, Kunqiu Chen, Lingming Zhang, Yuqun Zhang
Abstract
Long-horizon coding tasks contain nested subtasks. Typically, agents record interactions sequentially into a growing history, causing context inflation and performance degradation. Compressing or trimming the history without explicit subtask boundaries can discard information needed by unfinished subtasks. Conversely, recursive subagent delegation isolates completed subtasks through child agents by assigning every subtask boundary to delegation, which conflates subtask granularity with executing agent choice, adding execution and coordination overhead. We introduce Spine, an interface and runtime that align model context with subtask nesting and completion. An agent declares explicit subtask boundaries through work scopes, allowing the same agent to execute nested subtasks inline while reserving delegation for independent parallel tasks. The runtime organizes execution in SpineTree, a persistent tree recording subtask relations, local interactions, completion states, and returned results. Before each model call, it compiles the tree into model context just in time, retaining unfinished subtasks' interactions and replacing completed subtasks' details with their results, while preserving prefix-cache reuse. We integrate Spine into Codex to build SpineCodex. With GPT-5.6 Sol, compared with native Codex, SpineCodex improves SWE-Milestone Score by 11.8% while reducing total cost by 32.4%, improves the average ProgramBench test-pass rate by 17.3%, and improves Mean@5 on eight of the ten evaluated FrontierSWE tasks. On SWE-Milestone, it also achieves a comparable Score to Recursive Subagent at 73.7% lower total cost. We also integrate Spine into Pi through a reusable software package: with Gemini 3.8 Flash and Qwen 3.8 27B, it improves SWE-Milestone Score by 9.5-17.5% and the average ProgramBench test-pass rate by 4.3-8.7% compared with native Pi.