AI News

Anthropic is under pressure to add native AGENTS.md support to Claude Code after Shopify CEO Tobi Lütke threatened an internal ban. AGENTS.md is used by more than 60,000 open-source projects and multiple coding tools, while Claude Code relies on CLAUDE.md. Teams can bridge the formats with imports, symlinks, and automation, but those workarounds add maintenance costs in large monorepos. Anthropic says it is improving customization, without clearly promising native recursive support.

SSAKG 2.0 is an open-source package for representing ordered sequences as sparse graph structures and reconstructing them from incomplete, unordered context. Its new bit-level search algorithms and C-backed graph operations target lower memory and computational overhead. Experiments cover numerical, NLTK sentence, and mRNA sequences.

An inspection of the OpenAI Codex desktop application's cache revealed a 1.7GB runtime containing Python, Node.js, Poppler, Git, and LibreOffice. Codex document-plugin skills explain how to locate and use these bundled binaries, offering a look at the tooling behind its document workflows.

This AI Daily roundup covers Jimeng Studio's long-form AI film support, DeepSeek's reported RMB 475 million in seven-month revenue, Manus data recovery, Claude Cowork memory interoperability, OpenAI's reported Jalapeno chip, Stability AI funding, Grok Bot updates, and OpenRouter's asynchronous video API.

AI is moving into physical environments as robotics models become more reusable and hardware becomes cheaper. Yunzheng Li argues that long-term value will shift from robot bodies to intelligent models, real-world data, middleware, tools, and developer ecosystems. He also describes Agent-native development, where AI agents configure environments and run robotics demos. China’s manufacturing strengths and the United States’ model and software ecosystems may provide complementary foundations for global robotics platforms.

Coinbase CEO Brian Armstrong says AI agents can now trade equities on the platform, in addition to crypto and derivatives. Coinbase documentation says equity features are beginning to roll out for Agents through the Coinbase MCP, enabling stock research, sentiment analysis, performance tracking, and trade execution. The announcement marks a move toward “agentic finance,” but the supplied evidence does not independently confirm availability, geographic scope, trading limits, safeguards, or the full operational scope.

Robin Linacre explains how he used LLMs to build Letterpaths, a cursive-handwriting library, despite their inability to generate precise letter geometry or reliably improve joins from screenshots. By having models create editors, visual debugging tools, and font-building workflows, he established a fast human feedback loop for inspecting and correcting the output. The project suggests that LLMs can greatly accelerate software development when humans retain control over quality-critical decisions and use focused GUIs to diagnose problems.

David Ondrej details the agentic engineering setup he developed after more than 2,000 hours of AI-assisted coding. The workflow combines multi-model interfaces, terminal orchestration, persistent cloud agents on a VPS, reusable skills, isolated worktrees, guardrails, targeted reviews, and read-only production access. He also explains how he chooses models and subscriptions, and why human control remains important as manager agents begin coordinating worker agents.

A developer shares his move from Claude Code to Pi and explains how a small coding-agent core can support a highly customizable harness. AGENTS.md defines stable rules, Skills add capabilities on demand, and Packages provide orchestration, research, context, and side-thread controls. His workflow separates clarification, read-only scouting, implementation, and fresh review, with only one writer allowed per directory. He also warns that Pi has maximum privileges and no sandbox by default.

This technical tutorial explains Pi’s production agent loop by comparing its 743-line implementation with a simpler mini-agent. It covers nested loops, steering and follow-up queues, streaming, parallel execution, abort signals, and lifecycle hooks. The article also details the three checks before tool execution: tool lookup, TypeBox argument preparation and validation, and policy enforcement through beforeToolCall. Errors are returned to the model as tool results so it can correct itself and continue safely.

The final Oh My Pi tutorial explains how to turn Pi into a structured AI Agent workbench with project rules, reusable Skills, automation interfaces, review checkpoints, and stronger permission boundaries.

A hands-on comparison of five languages for a weather CLI finds Go offers the best overall balance for AI agents and MCP tools. Rust is faster but costs more time to build, Bun is pleasant but produces large binaries, and Zig remains difficult for production-ready tooling.