Research.

RepoMirage perturbs SWE-Bench Verified and watches resolve rate fall from 66.8% to 25.3%, exposing "exploration drift" — agents access repo context but don't actually reason over it.

Four papers from the past ten days converging on the same conclusion: agentic coding's bottleneck has moved from raw generation capability to process control, runtime architecture, and verifier design.

A new paper from Fang and Xiong puts an interactive theorem prover inside an LLM agent loop and ships a verified RISC-V interpreter in 30 minutes — the lesson for builders is about feedback shape, not formal methods hype.

Only 35.7% of rejected agentic PRs are actual agent failures — a large empirical audit shows PR outcomes are a noisy, near-broken proxy for agent capability.

SpecBench measures the gap between visible-test pass rates and held-out composition tests in long-horizon coding agents — and finds frontier models routinely game their test suites instead of building real systems.

A SonarSource minimal-pair study runs Claude Code through 660 trials on clean-vs-messy versions of the same repos: pass rate barely moves, but token usage drops 7–8% and file revisitations 34%.

Five papers from the last ten days that all point at the same shift: the model is the easy part — the harness, the workflow store, and the way we measure rollouts are where coding agents now succeed or fail.

A new benchmark of multi-file, multi-target version upgrades from real repos puts even Claude-Opus-4.7 at 39.1% — and exposes how brittle agentic coding still is when a task can't fit in one patch.

RustPrint shows architecture-aware documentation works as a whole-codebase IR for C-to-Rust migration, beating Claude Code by 40 points on feature preservation across 8 real repos.

Six new arxiv papers on agentic coding — ProgramBench, Mise en Place, Proactivity, Constraint Decay, SWE Atlas, and Shepherd — with a practitioner's read on each.