--- kind: trend trend_doc_id: 1317 granularity: day period_start: '2026-06-02T00:00:00' period_end: '2026-06-03T00:00:00' topics: - agentic software engineering - LLM training - code repair - terminal agents - multi-agent orchestration - developer tooling - agent trust run_id: materialize-outputs aliases: - recoleta-trend-1317 tags: - recoleta/trend - topic/agentic-software-engineering - topic/llm-training - topic/code-repair - topic/terminal-agents - topic/multi-agent-orchestration - topic/developer-tooling - topic/agent-trust language_code: en pass_output_id: 224 pass_kind: trend_synthesis --- # Evidence-rich control loops are the center of agentic coding work ## Overview The period treats large language model (LLM) agents as trainable and inspectable software systems. EvoTrainer, FLARE, and SPOQ carry the strongest evidence: better agents come from diagnostics, gated execution, and task structure, with human judgment present at planning and verification points. ## Findings ### Training harnesses and interaction traces Agent training work focused on the evidence available during training, not only on final task scores. EvoTrainer evolves policy branches together with the diagnostic code that reads rollouts, logs, configs, and code diffs. Its strongest reported gain is on repository-level software engineering: SWE-9B reaches 38.16 Avg@8 BC% versus 33.77 for a human-engineered reinforcement learning setup. The terminal-agent study reaches a related conclusion through supervised fine-tuning. DeepSeek-V3.2 has a lower standalone Terminal-Bench 2.0 score than Claude Opus 4.6, yet its traces train stronger Qwen3 students because they expose more inspect-act-verify behavior. Masking observation-command links cuts the Targeted Observation Ratio from 13.4% to 5.3% and drops Qwen3-32B performance from 20.6% to 13.8%. #### Sources - [EvoTrainer: Co-Evolving LLM Policies and Training Harnesses for Autonomous Agentic Reinforcement Learning](../Inbox/2026-06-02--evotrainer-co-evolving-llm-policies-and-training-harnesses-for-autonomous-agentic-reinforcement-learning.md): EvoTrainer summary, mechanism, and SWE-9B result against human-engineered RL. - [What Makes Interaction Trajectories Effective for Training Terminal Agents?](../Inbox/2026-06-02--what-makes-interaction-trajectories-effective-for-training-terminal-agents.md): Terminal-agent teacher comparison, trace properties, and TOR ablation results. ### Fine-grained code repair signals Code repair papers made debugging feedback more local and testable. FLARE adds line-level suspiciousness scores to execution feedback, then searches candidate edits over the top-k lines. With k=10, it improves Pass@1 across five base models on LiveCodeBench and BigCodeBench, while its diagnostic model alone reaches 67% Top-1 and 89% Top-10 localization accuracy on 100 LiveCodeBench tasks. Neural Change Prediction uses synthetic mutations and observed output changes to learn both directions of the edit-behavior link. Fine-tuned GPT-4.1 reaches 95% accuracy on CSS change prediction, 82.6% for Python change-location prediction with single mutations, and 95% for predicting output changes from a single Python code mutation. The practical value is clear: proposed repairs can be ranked by expected behavior and then checked by tests. #### Sources - [FLARE: Fine-Grained Diagnostic Feedback for LLM Code Refinement](../Inbox/2026-06-02--flare-fine-grained-diagnostic-feedback-for-llm-code-refinement.md): FLARE approach, benchmark gains, candidate search, and localization results. - [Neural Change Prediction: Relating Software Changes to Their Effects and Vice Versa](../Inbox/2026-06-02--neural-change-prediction-relating-software-changes-to-their-effects-and-vice-versa.md): Neural Change Prediction data construction and CSS/Python accuracy results. ### Multi-agent task orchestration with human gates SPOQ treats software work as a dependency graph. It splits an epic into atomic tasks, schedules independent tasks in parallel waves, and applies planning and code validation gates with a 95% aggregate threshold. Reported results include up to 14.3x speedup on unbounded synthetic graphs, 1.4x on a 2-slot real LLM backend, and a test pass-rate increase from 91.25% to 99.75% when dual validation is used. A broader synthesis paper gives the organizational reason for these gates. It cites 456,535 agent-authored pull requests across 61,453 repositories, but also notes that agent-authored pull requests are merged less often and tend to make fewer structural code changes. The human role described in this corpus is specification, review, governance, and escalation during ambiguous planning or risky changes. #### Sources - [SPOQ: Specialist Orchestrated Queuing for Multi-Agent Software Engineering](../Inbox/2026-06-02--spoq-specialist-orchestrated-queuing-for-multi-agent-software-engineering.md): SPOQ task graph, validation gates, human-as-agent design, and reported execution results. - [Human-AI Collaboration and the Transformation of Software Engineering Work](../Inbox/2026-06-02--human-ai-collaboration-and-the-transformation-of-software-engineering-work.md): Evidence on agent-authored pull requests and the proposed human competency model. ### Runtime cost, memory, and trust metadata Developer-facing tooling targeted the operating costs around coding agents. Cross-Lingual Token Arbitrage rewrites multilingual prompts locally before cloud dispatch. On OMH-Polyglot, prompt tokens fall by 34–47% while accuracy is held or improved across gpt-3.5-turbo, gpt-4o, and gemini-2.5-flash-lite, though dollar savings vary by backend because output-token pricing can offset input savings. Project Brain addresses repeated context setup with a small Markdown index that points to topic files. It has no benchmark, but the design is concrete: load the index eagerly, open detailed project notes only when relevant, and validate broken pointers or status drift. A separate agent-network paper adds a trust angle: capability ads should include reliability estimates, benchmark names, sample sizes, evaluation dates, version limits, expiry, tests, and reputation updates. #### Sources - [Cross-Lingual Token Arbitrage: Optimizing Code Agent Context Windows via Local LLM Preprocessing](../Inbox/2026-06-02--cross-lingual-token-arbitrage-optimizing-code-agent-context-windows-via-local-llm-preprocessing.md): Local prompt rewriting design and token, accuracy, and cost results. - [Project Brain – Persistent memory index for AI coding](../Inbox/2026-06-02--project-brain-persistent-memory-index-for-ai-coding.md): Project Brain memory layout, claimed token behavior, and lack of measured results. - [Capability Advertisement as a Market for Lemons: A Trust Layer for Heterogeneous Agent Networks](../Inbox/2026-06-02--capability-advertisement-as-a-market-for-lemons-a-trust-layer-for-heterogeneous-agent-networks.md): Trust Layer proposal for evidence-backed capability claims and drift checks.