Every processed story in chronological order, with the newest coverage first. Filter by tag, source, or score to drill in.
The findings show that agentic coding tools reward domain understanding over formal programming training, with non-engineers succeeding at roughly the same rate as software engineers — a direct signal about how these tools may reshape the labor market for knowledge workers.
The finding that 80.2% of agent-authored test patches lack meaningful assertions means that quality gates relying on test-file presence give a false signal of verification coverage in AI-generated code.
The finding that non-software occupations achieve success rates within 7 percentage points of software engineering on Claude Code's strictest metric suggests the tool's effectiveness is not limited to developers.
The study provides the first empirical baseline on how developers configure agentic coding tools across a large set of real-world repositories, establishing that `AGENTS.md` serves as a natural cross-tool starting point and that advanced configuration mechanisms remain largely underutilized.
The paper provides a concrete methodological foundation for characterizing SWE agent behavior in real repositories, turning raw trajectory data into disciplined, comparable behavioral profiles across models and task conditions.
Benchmark your agentic tooling against these metrics — 87% time reduction and 55% lower dissatisfaction — as the paper establishes a concrete empirical baseline for what autonomous end-to-end execution delivers over conversational search in real production settings.
A new repository in the agentic coding space raises questions about how context conditions affect benchmark reproducibility for coding agents.
Practitioners building Claude-based coding agents or prompt pipelines should prioritize rejection-logic prefixes like `/skeptic` and `L99` over additive "be more expert" instructions, which this study found produced no measurable reasoning improvement.
Practitioners and researchers evaluating AI coding agents can use SWE-chat's real-world interaction traces to benchmark agent reliability, study failure modes, and design interventions that address the security and code-survival gaps that curated benchmarks miss.
Practitioners can stop wasting time on hyped prompt codes like `GODMODE` and `BEASTMODE`, and instead focus on the 7 empirically validated codes — especially `/skeptic` and `L99` — to meaningfully change Claude's reasoning behavior rather than just its tone.