Scott Fryxell describes an AI coding harness that lets Cursor, Claude, and Pi share instructions, skills, and an AGENTS.md file across terminal-based interfaces. He argues that this shared operating environment matters more than any individual model and says it has reduced his concern about switching from Cursor to Codex. His current setup uses two $20 subscriptions, Cursor and Claude, with Pi added for other models. Fryxell uses DeepSeek-V4-Flash-0731 for maintenance and simpler work, reserving frontier models and Anthropic API budget for complex features, refactors, and promotion tasks. His workflow separates work into explorer, planner, worker, critic, and promoter roles. Planning produces an explicit directed acyclic graph of tasks; a worker implements nodes, a critic reviews and may send work back for revision, and a promoter prepares communications about completed work. He says this cut his use of Fable by 75 percent in intensive work. The harness contains scripts, prompts, plans, skills, extensions, artifacts, and checked-out work repositories. It can control his poster-making web app through headless Chrome, including an npm command that generates an animation from an MP4. Fryxell says a government ban on Fable and signals of industry control made diversifying model access urgent, leading him to rely more heavily on Pi and Chinese models.

Microsoft released Agent Lightning v1.0.1, the first official release of the Agent Lightning Skill for coding agents. The skill is designed to help coding agents optimize other AI agents by working from an editable agent and a benchmark. Agent Lightning guides measured, systematic iteration across prompts, tools, workflows, models, and reasoning settings. It balances accuracy, cost, latency, and reliability during those improvements. The skill can be installed for Claude Code, Codex, or GitHub Copilot with the command `gh skill install microsoft/agent-lightning agent-lightning --agent <agent>`. Version 1.0.1 also strengthens continuous integration, packaging, release automation, documentation, and benchmark reporting.

GPT-5.6 utilized a prompt to bridge a 30-year gap in convex optimization research. This advancement builds on OpenAI's recent proof announcement related to convex optimization.

Lucebox is an optimization hub for hand-tuned LLM inference, specifically designed for individual consumer hardware. It features kernels, speculative decoding, and quantization tailored for each target, with the first megakernel for hybrid DeltaNet/Attention LLMs achieving 1.87 tokens per joule on a 2020 GPU.