mlsanityData to 14 Oct 2026

The Handoff Tax: Continuing Non-Native Trajectories in LLM Agents

arXiv 2608.24358 ↗First posted 25 Aug 2026cs.AIcited by 6

On the map: LLM agents and tool use → Software engineering coding agents

Abstract (arXiv)

Coding agents perform long-running tasks spanning dozens of model calls, tool uses, and code edits. As these runs unfold, users face a practical cost-quality trade-off: escalating to a stronger model when a cheaper one struggles, or downshifting once the hard reasoning is complete. Each switch requires the receiver to continue a non-native trajectory produced by another model. We study how this handoff affects quality and cost, and how varying the trajectory information inherited by the receiver changes the outcome. Using pairs of low-cost, low-capability (LC) and high-cost, high-capability (HC) models from the Claude and GPT families, we vary handoff direction, timing, and interface, comparing full-trajectory transfer, compaction, and trajectory removal while preserving the repository state. Across both model families, full-trajectory escalation recovers less than half of the LC-to-HC quality gap while incurring a substantial cost premium. We term this cost-quality penalty the handoff tax. By contrast, downshift offers a favorable cost-quality point. Interestingly, the preferred interface also reverses with direction: reducing LC-model trajectory information improves escalation quality, whereas removing the HC-model trajectory reduces downshift quality.

Key facts. Cited by 6 papers in the map (data through 14 Oct 2026). Placed in Software engineering coding agents, one of 16 directions in LLM agents and tool use. Builds on 27 of its 34 references that are in the map. First posted on arXiv 25 Aug 2026.

Builds on

27 of 34 references.

Closest ancestors · 3

Same direction, newest first

  1. 2026SWE-Router: Routing in Multi-turn Agentic Software Engineering Tasks30 Jun
  2. 2026Handoff Debt: The Rediscovery Cost When Coding Agents Take Over Interrupted Tasks1 Jun
  3. 2024Executable Code Actions Elicit Better LLM Agents1 Feb

Other references · 24

  1. 2021Training Verifiers to Solve Math Word Problemscited by 10,109
  2. 2022ReAct: Synergizing Reasoning and Acting in Language Modelscited by 8,695
  3. 2021Evaluating Large Language Models Trained on Codecited by 8,320
  4. 2023Toolformer: Language Models Can Teach Themselves to Use Toolscited by 4,182
  5. 2023SWE-bench: Can Language Models Resolve Real-World GitHub Issues?cited by 3,373
  6. 2024LiveCodeBench: Holistic and Contamination Free Evaluation of Large Language Models for Codecited by 2,004
  7. 2023ToolLLM: Facilitating Large Language Models to Master 16000+ Real-world APIscited by 1,947
  8. 2018A Call for Clarity in Reporting BLEU Scorescited by 1,799
  9. 2023Gorilla: Large Language Model Connected with Massive APIscited by 1,332
  10. 2018Spider: A Large-Scale Human-Labeled Dataset for Complex and Cross-Domain Semantic Parsing and Text-to-SQL Taskcited by 1,139
  11. 2023FrugalGPT: How to Use Large Language Models While Reducing Cost and Improving Performancecited by 853
  12. 2024RouteLLM: Learning to Route LLMs with Preference Datacited by 682
12 more
  1. 2025BrowseComp: A Simple Yet Challenging Benchmark for Browsing Agentscited by 647
  2. 2025LLMs Get Lost In Multi-Turn Conversationcited by 429
  3. 2024Hybrid LLM: Cost-Efficient and Quality-Aware Query Routingcited by 372
  4. 2020ToTTo: A Controlled Table-To-Text Generation Datasetcited by 268
  5. 2023Large Language Model Cascades with Mixture of Thoughts Representations for Cost-efficient Reasoningcited by 143
  6. 2020Semantic Evaluation for Text-to-SQL with Distilled Test Suitescited by 126
  7. 2026MTRouter: Cost-Aware Multi-Turn LLM Routing with History-Model Joint Embeddingscited by 10
  8. 2025C3PO: Optimized Large Language Model Cascades with Probabilistic Cost Constraints for Reasoningcited by 8
  9. 2026Evaluating Performance Drift from Model Switching in Multi-Turn LLM Systemscited by 6
  10. 2026From Myopic Selection to Long-Horizon Awareness: Sequential LLM Routing for Multi-Turn Dialoguecited by 4
  11. 2026UCCI: Calibrated Uncertainty for Cost-Optimal LLM Cascade Routingcited by 4
  12. 2026R2V Agent: Teaching SLMs When to Ask for Helpcited by 2

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 6.

6 papers.

  1. 2026False Floors: LLM Safety Routing Evaluations Break Under Distribution Shiftcited by 0
  2. 2026Attention Tax, Handoff Tax: A Stylised Model of When Multi-Agent LLM Systems Helpcited by 0
  3. 2026Humanize MLE Agents: Multi-Agent Session Alternation Beyond Single-Agent Limitscited by 0
  4. 2026Handoff Curves: Initial Selection, Later Handoff, and Supervision Targets in Multi-turn LLM Routingcited by 0
  5. 2026Measuring and Reducing Unusable Replies of LLM Agentscited by 0
  6. 2026New Model, Old Habits: Behavior Transfer under Model Swapping in Multi-Session LLM Agentscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026TACIT-Switch: Cost-Aware Model Escalation for LLM Agents from Censored Supervision
  2. 2025Reducing Cost of LLM Agents with Trajectory Reduction
  3. 2026TraceLab: Characterizing Coding Agent Workloads for LLM Serving
  4. 2026Beyond Token Savings: A Systematic Study of Context Compression in LLM Agents
  5. 2026The Replay Gap: Static Evaluation of Model Switching in LLM Agents Scores the Wrong World
  6. 2026A Systematic Evaluation of Trajectory Data Curation for LoRA Fine-Tuning of Code Agents
  7. 2026How Fast Do Agents Rot? An Empirical Study of Long-Horizon Degradation in LLM Agents for Production Decision-Making
  8. 2026Agents Don't Paginate: First-Chunk Selection for LLM Tool Responses