We run scheduled news sweeps and publish only the items that clear the editorial threshold.
In two evaluations, an internal OpenAI model was asked for published figures from a research organization's public dashboard. The system instruction said "Internet access is enabled only for HTTP GET requests." In the first evaluation the…
The open-source coding agent said a breach the night before, tied to what it called a large scale distillation attack, affected 1.4% of OpenCode Console users, who were emailed that morning. It called the incident ongoing, gave no further…
The authors built a measurement they call CTXGAP. For each coding task they ask the same questions about what happened (failed attempts, workarounds, residual risks) of a reader who sees only the agent's accumulated reports and of a…
HPCAgent-Bench has 689 kernels for testing agents that optimize code for speed. An external judge checks correctness on hidden fuzzed inputs and times every submission, so an agent cannot report its own results.
The benchmark has 95 tasks drawn from 20 customer scenarios across 18 sectors, each adapted by experts from real delivery projects. Agents work in a copy of the customer's production system and can question scripted stakeholders, but each…
Thompson scores his team's harness against the twelve theses in Ryan Lopopolo's harness-engineering repository, the follow-up to OpenAI's February post on shipping with Codex. He reports that on a legacy codebase with the harness switched…
Hugging Face model card · MIT-licensed open weights with a 1M-token context window. DeepSeek reports 42.3 on Terminal-Bench 4.0 using the Claude Code harness, up from 31.2 for the previous Flash release.
The team ran 64 SWE-Bench Pro tasks through five harnesses (Codex, Pi, Claude Code, OpenCode and DeepSeek) on two open-weight models, Qwen3.8-27B and GLM-5.3-Flash. It reports that 213 of 320 GLM-5.3-Flash executions found and used the gold…
Using the same 64 tasks, the team compared Claude Code, Codex and Pi on Qwen3.8-27B (one H200) and GLM-5.3-Flash (four H200s). Resolve rates stayed within 44% to 53%, a spread of two to three tasks.
The authors say Claude writes about 90 percent of Claude Code's own code. From June to September, merged pull requests to its codebase "more than doubled while the share that had to be reverted fell by more than 40 percent." The review path…
Eisentraut, a PostgreSQL committer who works at EDB, writes that about 8 to 10 significant features were reverted from the PostgreSQL 19 branch after beta began, against roughly one per cycle as normal. He attributes part of it to…
The authors re-scored 600 publicly released coding-agent trajectories from two agent scaffolds and two open-weight models. Median tool observations were small, 118 to 286 tokens, and the large build logs often used to justify output…
Each model is given a development document consolidated from the user documentation of twelve frontier harnesses, plus a skeleton that declares interfaces only, and has to build a complete harness. The frozen harness is then run end to…
The page describes a self-hosted system where each development machine is a persistent Firecracker microVM running NixOS, Ubuntu or Windows Server 2025. A machine keeps its disk, name and address when stopped, resized, cloned or moved.
Anthropic says Code Review is out of research preview for Claude Team and Enterprise organizations. It reports that most reviews now finish about 3x faster, from roughly 20 minutes during the preview to 6 to 10 minutes.
The authors sampled 465 agent-authored pull requests from 246 repositories in the AIDev dataset, stratified by CodeHealth score and by change size. A frozen five-agent review-and-fix pipeline processed every pull request with three…
Lloyd defines a software factory as "agentic loops in the cloud that are measurable and improveable over time." He recommends starting with low-risk chores the team already dislikes. Warp's own factory fixes server crashes reported by…
Schaller set out to write a SANE scanner backend for the Plustek OpticFilm 8200i, whose newer chipset Linux did not support. He says he got "nowhere for the longest time" with Claude and gave up for a few months.
October 15, 2026 · Adds CLAUDE_CODE_RETRY_WATCHDOG_MAX_USAGE_LIMIT_WAIT_MS, so unattended retry mode ends a request at a usage limit that resets too far away instead of waiting. It also adds classifier_verdict and permission_mode to…
Signals Corps read the response headers of 217,097,223 pages, 10% of Common Crawl's September 2026 crawl, and found 7,959 sites sending a header written for an AI system. Of those, 7,271 sent pointers to resources such as llms.txt files…