Reading List

Week 1 — From LLMs to Coding Agents

  1. The Llama 3 Herd of Models — Grattafiori et al., 2024 arXiv
  2. SWE-agent: Agent-Computer Interfaces Enable Automated Software Engineering — Yang et al., 2024 arXiv

Question: What does a foundation model need to become a useful software-engineering agent?

Week 2 — Agent Architecture

  1. ReAct: Synergizing Reasoning and Acting in Language Models — Yao et al., 2023 arXiv
  2. Agentless: Demystifying LLM-based Software Engineering Agents — Xia et al., 2024 arXiv

Question: Does sophisticated agent architecture actually improve software engineering?

Week 3 — Repository Understanding

  1. RepoCoder: Repository-Level Code Completion Through Iterative Retrieval and Generation — Zhang et al., EMNLP 2023 ACL Anthology
  2. AutoCodeRover: Autonomous Program Improvement — Zhang et al., 2024 arXiv

Question: How does an agent determine which parts of a repository it needs to understand?

Week 4 — Context Engineering

  1. RepoGraph: Enhancing AI Software Engineering with Repository-level Code Graphs — Ouyang et al., ICLR 2025 ICLR paper arXiv
  2. Context as a Tool: Context Management for Long-Horizon SWE-Agents — Liu et al., Findings of ACL 2026 ACL Anthology

Question: Is context management becoming a first-class capability of coding agents?

Week 5 — Tools & Action Spaces

  1. CodeAct: Executable Code Actions Elicit Better LLM Agents — Wang et al., 2024 arXiv
  2. CODESTRUCT: Code Agents over Structured Action Spaces — Kim et al., ACL 2026 ACL Anthology

Question: Should an agent interact with a codebase through text, executable actions, or structured program representations?

Week 6 — Planning, Search & Long-Horizon Work

  1. CodePlan: Repository-level Coding using LLMs and Planning — Bairi et al., FSE 2024 Microsoft Research
  2. SWE-Dev: Building Software Engineering Agents with Training and Inference Scaling — Findings of ACL 2025 ACL Anthology

Question: How should agents plan and execute work that spans many steps and files?

Week 7 — Verification & Feedback

  1. Agentic Rubrics as Contextual Verifiers for SWE Agents — ACL 2026 ACL Anthology
  2. RLEF: Grounding Code LLMs in Execution Feedback with Reinforcement Learning — Gehring et al., ICML 2025 arXiv

Question: How does an agent determine whether its solution is correct, and how should feedback change its behavior?

Week 8 — What Should We Measure?

  1. SWE-bench: Can Language Models Resolve Real-World GitHub Issues? — Jimenez et al., ICLR 2024 arXiv

Plus SWE-bench Verified, the human-validated subset: OpenAI — SWE-bench Verified

  1. OmniCode: A Benchmark for Evaluating Software Development Agents — Sonwane et al., Findings of ACL 2026 ACL Anthology

Question: What do coding-agent benchmarks actually measure, and what do they systematically miss?

Week 9 — Harness Engineering

  1. Harness Engineering: Anatomy, Architecture, and Evolution of Coding Agents — A Source-Code Study of Eleven Systems — Barbaste, Darrigol, Vu & Wiltberger, 2026 arXiv
  1. Agentic Harness Engineering: Observability-Driven Automatic Evolution of Coding-Agent Harnesses — Lin et al., 2026 arXiv

Question: If the model is only one component of an agent, what should the surrounding runtime—the harness—actually do?

Week 10 — Where Coding Agents Are Going

  1. From Laboratory to Real-World Applications: Benchmarking Agentic Code Reasoning at the Repository Level — Li, Su & Lyu, ACL 2026 ACL Anthology
  2. Contract-Coding: Towards Repo-Level Generation via Structured Symbolic Paradigm — 2026 ACL Anthology

Question: What abstractions will be required for agents to progress from patch generation to genuine repository-scale software engineering?