#Coding

AINous Research says its evaluation uses an index of four benchmarks, including its newly introduced Hermes Bench. The other three are Terminal-Bench 4.0, Terminal-Bench-Science, and SkillsBench.

AIGitHub reports that Git events on the platform rose from 218.2 billion to 473.3 billion per month between September 2025 and August 2026. It says agent workloads push write throughput and merge contention beyond what its current replica-based architecture handles well, so it is separating durable storage from compute while GitHub keeps running. The article states internal benchmarks reached up to 35 times higher write throughput.

Why it matters: The post links rising Git event volume to specific architectural bottlenecks, showing why agent workloads strain write paths and how GitHub plans to separate storage from compute.

AIGemini CLI v0.63.0 adds a retry progress indicator during connection recovery and fixes an infinite authentication loop caused by file contention, headless keyring issues, and supervisor state drops. The release also bounds tool output size and cleans up temporary directories when background shell execution exits, alongside fixes for MCP enablement config handling and stdin restoration after capability detection.

AIBoris Cherny says he used Opus 5.5 with Lean to formally verify the Claude Agent SDK, with a couple of short prompts producing 16 PRs fixing bugs and race conditions. He also reports that TLA+ works well, sometimes combined with Lean to find data flow, concurrency, and state management issues. The post links to his actual prompts as another example.

AIGoogle Research will demo ScientistTwo, an autonomous multi-agent framework, at the COLM booth (#107) today at 2:00 PM PT. The framework analyzes research papers, identifies their limitations, and produces verified codebases. The accompanying paper is available on arXiv (2609.19644).

AIAnthropic is expanding its Cyber Verification Program to give verified security professionals broader access to its most capable models. Through the program, they can use Claude Mythos 5.1, Opus 5.5, and Sonnet 5.5 with safeguards designed for defensive work. New tiers will also allow authorized offensive work such as penetration testing and red-teaming.

AIClaude Code v2.1.292 adds a --marketplace option to claude plugin install, which adds the marketplace if needed and then installs the plugin from it. The release also adds an effort parameter to the Agent tool and fixes several security issues, including permission prompts bypassed for network (UNC) file reads and a sandboxed read path that could return files outside approved access.

AIAnthropic's ClaudeDevs account says Pro or Max plan subscribers on September 23 can still claim a one-time bonus credit for cloud sessions by running /claim-credit in Claude Code by October 7 at 11:59pm PT. Cloud sessions draw on this credit first before counting toward plan limits, and the credit expires November 4.

AIAnthropic's ClaudeDevs says Claude Code cloud sessions run each task on a fresh VM, so users can start several at once. The sessions keep running after the user closes their laptop. A field guide covers seven suitable workflows and how to connect GitHub.

AIGitHub has released ReviewBench, a new open benchmark for evaluating AI code review agents, shaped by analysis of 103.9M GitHub pull requests. It contains 219 PRs across 19 languages, and developers can bring their own code review agent, evaluate it, and submit their results.

AIMistral Large 4 solved 18 of 19 challenges in a CTF speedrun, with tool calls and solve times drawn from actual runs. The post frames the model as efficient at reasoning over diverse complex challenges compared with other models.

AIA developer used Grok Bot to build two Slack team bots, sandcastle for release management and poteto for engineering, automating their release and QA process. Sandcastle DMs contributors PR links, kicks off builds, and runs a fuzz swarm of 10+ Grok 4.7 xhigh agents, while poteto triages and fixes issues via Cursor Projects.

AIThe xai-cookbook example uses Grok to generate a Tailwind React component from a dropped-in screenshot. The page then screenshots its own render, overlays it on the original, and feeds the mismatches back so Grok can refine the layout, raising the match from 60% to 77% in one run.

AIAi2 applied scientist Shashank Gupta says he will attend COLM 2026 from Tuesday through Friday and invites people to talk with the Ai2 Agents post-training team. Topics include post-training for coding and long-horizon agents, such as agentic RL, OPD, data and infrastructure, and multi-agent training, plus opportunities at Ai2. He lists an Ai2 booth session Tuesday 1:30–3pm and an Ai2 mixer Tuesday 6–9pm.

AILauren Tan describes using Grok Bot to connect apps like Slack, email, and CRM so it can gather context and decide what work to do. She also explains scheduled and event-triggered routines, plus handing engineering tasks to Cursor cloud agents, with a daily feedback-to-merge loop example.

AIThariq argues that working at a higher level of abstraction has always required understanding the levels beneath it. He says coding agents do not change this requirement.

AICodex now offers an "Approve for me" permission mode, which automatically reviews actions instead of requiring manual approval. To enable it, open the permissions menu below the composer and select "Approve for me."

AIAccording to Gergely Orosz, GitHub now records more AI-generated pull requests per month than human-written ones. He says the volume of AI-generated PRs is about three times what it was in December 2023.

AIReplit now lets users start a new chat and ask it to locate existing projects, add features, or use one project as a reference for another. The assistant can read project files and run edits as background tasks, with links to review changes and continue work.

AIGoogle Flow Music lets creators build custom instruments or effects from natural language, and Spaces can now be exported as VST3/AU plugins. These plugins run inside producers' Digital Audio Workstations, so tools can fit existing production workflows. The source gives producer Khris Riddick-Tynes's "No Chaser" plugin as an example for checking instrumentals and vocals.

AIOpenAI's Codex "Approve for me" mode uses a separate Auto-review agent to check actions needing approval, such as running commands outside the sandbox or accessing extra files and network resources. It reduces approval prompts during long tasks while keeping sandbox protections, and it is now free with ChatGPT sign-in without drawing from plan usage.

AIAt a recent AI workshop, switching the desktop app from ChatGPT to Codex drew the loudest reaction of the day from subject-matter experts who use ChatGPT daily. Allie K. Miller argues that enterprise AI knowledge is overestimated, suggesting AI training for SMB and enterprise teams is a lucrative opportunity.

AIDex Horthy reports that /show-me has been incorporated into the official Amp Code plugins. A linked GitHub pull request in the ampcode/official-plugins repository is referenced as the relevant change.

AIMistral AI's Mistral Large 4 is listed on OpenRouter as a frontier multimodal model accepting text and image input. The listing says it is built for reasoning, coding, and agentic workloads and offers a 1M-token context window. The feed excerpt is truncated, so further details such as pricing or availability are not confirmed here.

AIOn human evaluation, the model outperforms GLM 5.3 on STEM, CAD, and finance tasks and performs on par on agentic coding. The post is part 5 of a thread, so the model's name and other details come from earlier posts not included here.

AIMistral's ML4 model matches the best open-weight models on DeepSWE, AutomationBench, and AA-Briefcase, and reaches state-of-the-art results on finance and legal workflows and complex multimodal grounding benchmarks. The post says it can navigate terminal workflows, work across spreadsheets, slides, and PDFs, and reason over scientific and multimodal tasks.

AIUber's engineers describe migrating 600,000 JUnit 4 tests covering 15 million lines of code to JUnit 5, which the post author says was impractical by manual means. The author says AI now makes such a large migration feasible, and points readers to Uber's engineering blog for the details.

AIThe Darwin Gödel Machine, a coding agent from Sakana and Jeff Clune's lab, modified its own codebase over roughly eighty iterations without supervision. Its additions included better file viewing, patch validation before submitting fixes, generating and ranking several candidate solutions, and keeping a history of failed attempts. These changes raised its score from 20 to 50 percent on SWE-bench and from 14 to 31 percent on Polyglot.

AIOpenCode says its free tier will keep offering Space Bunny for a few more days after the original free period ends. The post also notes that the model is being tuned before its reveal.

AIOpenAI has made Auto-review free for all users signed in through a ChatGPT account, and it does not draw usage from their plan. Auto-review uses a second agent to check the primary agent's actions, blocking high-risk moves and actions that drift from user intent, so long tasks can run without constant approval prompts. It can be enabled under settings > permissions > auto-review.

AITibo of OpenAI advises users to switch from full-access to auto-review mode. He says the change is no longer a trade-off and offers greater peace of mind.

AIReflection announced Beam, a text-only 501B-total, 23B-active MoE model for coding, agentic, and scientific work, trained from scratch with full weights under Apache 2.0 promised this month. Self-reported results include 80.9 on SWE-bench Verified and 3–4x the inference efficiency of GLM 5.2, while the roundup notes that GLM 5.3, Kimi K3, Qwen 3.8 Max, and DeepSeek V4.1 Flash are generally ahead.

AIThariq (@trq212) explains how "local hands" is being implemented in Claude Code, a pattern where Claude runs in the cloud but can access files on the user's machine. The same approach is also coming to Cowork.

AIHarrison Chase, founder of LangChain, endorsed a post on harnesses with the brief comment "Good take on harnesses." The post, from @zeeg, argues that general coding harnesses like Codex will be superseded by specialized ones and that local models will handle most daily tasks within five years.

AIMastra has released Agent Controller in general availability, a runtime that hosts long-running agent sessions around the agent loop. The team says it was first built for Mastra Code and expanded to support Mastra Factory, which runs many concurrent sessions, and that memory usage in long-running Mastra Code processes dropped from 2–20 GB to 300–750 MB after optimizing UI state snapshots.

Why it matters: The post explains how the controller evolved from one developer's session to many concurrent sessions, with measured memory and storage changes useful to engineers building multi-user agent apps.

AIComcast and Booz Allen used Claude Mythos Preview to find vulnerabilities that arise from interactions across code, configuration, and deployment rather than single-file bugs. Comcast identified a critical authentication flaw across 258 systems and about 170 million lines of code before any exploitation was observed. Booz Allen reported that one analyst reviewed eight production systems across 138 repositories in twelve days, a review its team estimated would have taken several months without the model.

Why it matters: The case studies show how security teams validate and remediate model-found exploit chains, a workflow relevant to anyone managing large codebases.

AIAnthropic's Claude Code v2.1.291 release fixes a regression in 2.1.290 where cloud sessions could drop answers to permission prompts. It also fixes a regression in 2.1.288 where a session's last messages could be lost when quitting.

AIThariq says planning with HTML is much more token efficient than generating raw HTML. The model does not need to recreate components or logic for common elements such as state machines, diagrams, and code snippets. Background from the quoted post says he is building a Claude Code skill that generates HTML plans, with linting to reduce common failures.

AIZhipu GLM's GLM-5.3 is now available on Amazon Bedrock, offering enterprises coding and agentic capabilities. The post points readers to the Amazon Bedrock model card for GLM-5.3 for getting-started details.

AIReflection AI, an Nvidia-backed startup, released Beam, its first open-weight large model, aimed at coding and agent tasks. The company says Beam is comparable to Z.ai's GLM-5.2 and is approaching Qwen3.8-Max on coding and agent work. Beam has 501 billion total parameters, with 23 billion activated per task in a sparse architecture.