Coding

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.

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.

AIDex Horthy says a new Claude Code skill resembles the outline skill HumanLayer bundles, which mixes a /show-me command with planning language. He praises its HTML-based questions, interactive diagrams, and a pipeline that sends questions to the clipboard.

AIPuck users can now run several separate conversations at once, with each getting its own Puck, by pressing + to start one and clicking a conversation's name to switch. Conversations untouched for three days move into an Inactive list to keep the main list short.

AIThe author says Beam, a 501B-parameter open model from Reflection AI, comes close to GLM 5.2 in capability but trails GLM 5.3, Kimi K3, and DeepSeek V4.1 Flash in several areas. The author attributes Beam's competitiveness mainly to inference efficiency, with inference compute at roughly one-third to one-quarter of GLM 5.2's.

AINVIDIA has published Nemotron-Labs-3-Competitive-Coding on Hugging Face, a competitive-programming specialist model built on Nemotron-3-Ultra. The model is available in the NVIDIA-Nemotron-Labs-3-Competitive-Coding-550B-A55B-NVFP4 repository, indicating a 550B-parameter total size with 55B active parameters in NVFP4 format.

AICline says its CLI can be installed globally with npm i -g cline, and it has also released a new Desktop app. The company notes several free model promotions are available to try the Desktop app.

AIOllama says it is excited about a new open model from Reflection AI and that more US models are coming to Ollama. Reflection AI's Beam is described as an agentic open model with 501B total parameters and 23B active, with full weights due this month.

AIReflection AI has introduced Beam, an open agentic model with 501B total parameters and 23B active, trained end-to-end from scratch. The company says Beam advances the Western open frontier on coding and agentic tasks, with full weights set for release this month.

AICognition published a blog post about memory and dreaming, linking readers to the full article at devin.ai. The post itself gives no further details, so its specific claims cannot be summarized from this source alone.

AIAdo Kukic, an Anthropic-affiliated account, posted an enthusiastic one-line message saying Go plus Opus 5.5 means "shipping." The post gives no details on features, benchmarks, or specific use cases.

AILewis Tunstall reports that when Claude and Codex review each other's PRs, Claude's reviews are packed with blockers and nitpicks, while Codex usually approves with LGTM. He suggests this could mean Opus 5.5 is the strongest coder right now, or that GPT 6.1 Sol is simply more lenient.

AIDex Horthy argues there will always be an advantage in reviewing AI output, since unreviewed work tends toward generic "AI slop." He adds that the reviewed object may not always be code, and speculates that once models far surpass humans, human interference could make results worse.

AIArtist Leo Villareal used Gemini and Google Antigravity to build the authoring software behind his light sculpture "Buckyball," which debuted in New York City this summer. Google Cloud's post frames the project as a case study of AI tools supporting creative work. The source provides no further technical details about the software or results.

AICursor's SDK now supports steering and system prompts for local TypeScript agents. Background subagent results are returned locally in TypeScript and Python.

AIGitHub has released ReviewBench, an open benchmark for evaluating AI code review agents on 219 public pull requests across 19 languages. The benchmark reports grounded and augmented precision, recall, and F1 metrics, and its dataset, rubric, and judge are publicly available. GitHub says ReviewBench predicted the direction of a Copilot code review ensemble experiment's production results before A/B testing.

Why it matters: The post explains how ReviewBench was built and validated, and reports an offline-to-production comparison that shows how well a benchmark predicts real experiment outcomes.

AIFireworks AI has made DeepSeek V4.1 Flash available for training on its Dedicated Training API and Managed Training surfaces. The post positions the model as a strong base for agentic coding, terminal automation, and tool use, and notes it is cost-efficient to serve.

AIReplit shipped a weekly update letting users build with GPT-6.1 Sol and Claude Sonnet 5.5, along with an Ask agent integration with Jev. The release also includes an updated Settings UI and enterprise Workplace controls for company-wide rules and controlled exceptions. Full details are in the Replit changelog.

AICloudflare announced 46 products and updates during Birthday Week 2026, including the cf CLI for the entire Cloudflare API and EmDash, an open-source Astro-based serverless CMS whose plugins run in isolated Worker sandboxes. The company also said it plans to become a public certificate authority that issues free Merkle Tree Certificates for post-quantum authentication.

AITogether Link connects coding agents such as Claude Code, Codex, OpenCode, and Pi to open models on Together AI, which the company says cuts spend by over 50%. Setup takes one command, and its "Auto" mode routes each session's first task to a fast low-cost model or a frontier model, with a per-session tracker comparing costs against Opus 5.5.

AIOpenAI researchers' daily coding-agent usage, valued at API prices, rose from under $1 in January 2026 to $601 for the median researcher by mid-August. The 90th-percentile researcher reached over $7,000 per day, and both groups show doubling times of roughly one month. Epoch notes these are API-list values, not OpenAI's internal costs.

Why it matters: The figures show internal coding-agent usage growing fast enough to matter for research cost, though they measure API-list value rather than OpenAI's actual spending.

AIBoris Power, who is linked to OpenAI, posted "Time to 🚢 🚢 🚢!" to signal an imminent round of shipments. Quoted background from @thsottiaux says that over the next 28 days the team will ship one clear improvement relevant to most Codex and work users each day, or a full reset.

AIVercel completed migrating Turborepo from Go to Rust, which Rauch says was chosen for better low-level OS access despite controversial returns on human migration costs. He argues that what is best for humans is no longer necessarily best for business now that agents are writing code, and suggests Rust may not be the final toolchain.

AIJerry Liu says ChatGPT/Codex currently has the best agent interface for deep work, unifying coding and knowledge work in one place with forking support that Claude's app lacks. He still prefers Claude Code CLI as close to the best a CLI can be, and uses Opus 5.5 mainly through it for product demos, while noting a GUI is sometimes nicer.

AIThe poster calls GPT-6.1-Sol their go-to model for both work and personal use, saying it is "all you need." Citing Arena's real-usage cost-per-task data, they claim users should get about 5x more usage from GPT-6.1-Sol than from Astra.