// edition · 2026-06-21
15 stories on AI dev tools, agents, and the coding stack, curated from the day's RSS haul by Agentic Dev's pipeline.
Anthropic's Claude Code supports a feature called dynamic workflows, in which Claude generates custom JavaScript orchestration files to spawn and coordinate parallel sub-agents, each with isolated context windows and configurable models, to handle complex multi-step tasks.
A developer restructured Claude Code configuration files across a monorepo, replacing a single root AGENTS.md with a hierarchy of module-specific files capped at 120 lines each. Augment Code's study of AGENTS.md files found that overloaded configuration files degraded output quality below the bas...
A tutorial published on Dev.to covers how to set up Anthropic's Claude API with Node.js and TypeScript using the `@anthropic-ai/sdk` package, including installation, authentication, streaming, tool use, prompt caching, and error handling.
Cisco Foundation AI released a system called FAPO (Fully Automated Prompt Optimization) that identifies failures in multi-step AI prompt pipelines by evaluating each step independently, pinpointing the specific prompt causing errors rather than treating the pipeline as a single unit.
Contact forms generated by AI coding tools such as Cursor, Bolt.new, and Claude Code commonly fail to deliver submissions due to five recurring issues: missing action URLs, incorrect HTTP methods, absent name attributes on input fields, no spam protection, and no mechanism to verify delivery.
Anthropic's official Go SDK (`github.com/anthropics/anthropic-sdk-go`) allows developers to access the Claude API in under 20 lines of code. The SDK supports streaming, tool use, prompt caching, and error handling, with authentication via the `ANTHROPIC_API_KEY` environment variable.
Anthropic's Claude Cowork is a desktop app feature that executes multi-step tasks on a user's local files and applications, available on paid plans ranging from $17–$200/month. It accesses local files, connected apps like Slack, and live web browsing via Chrome to complete tasks such as file orga...
Project Valkey used an AI agent to automate bug fix backporting for its 9.1 release, handling cherry-picking, CI pipeline runs, and merge conflict resolution across older branches. The effort replaced a manual process that maintainers said consumed hours of engineering time per release cycle.
A developer describes accumulating over 2,000 AI conversations across ChatGPT, Claude, and other tools, and outlines methods for exporting, searching, and extracting useful information from stored chat histories.
An automated SEO system that improves calculator pages daily since May 11 made second-pass content updates to three pages (P/E Ratio, SaaS Valuation, Book Value) roughly 35 days after the first pass. Google Search Console data showed the P/E Ratio page accumulated 711 impressions and 2 clicks ove...
EGC (Extended Global Context) is a local runtime that adds persistent memory to AI coding tools via two MCP servers, storing session state, decisions, and preferences in Markdown files at ~/.egc/state/. It allows AI agents to reload project context automatically at the start of each new session.
A developer released "proven-python," an MIT-licensed Claude Code skill that requires AI coding agents to follow test-driven development, pass mypy --strict type checking, and clear ruff and pytest before marking Python code complete. It is available on GitHub at shanwije/proven-python and works ...
Google shut down Gemini CLI on June 18, replacing the open-source tool — which had over 100,000 GitHub stars — with a closed-source Go binary called Antigravity CLI. The new tool lacks full feature parity at launch, and personal-account users lost access to the old CLI without a grace period.
Anthropic pulled its Fable 5 and Mythos 5 models offline on June 12 to comply with a U.S. export-control directive barring foreign nationals from accessing them. Z.ai released GLM-5.2, an MIT-licensed open-weight model with a one-million-token context window, which Arena's Agent leaderboard ranke...
A developer published a framework arguing that AI-assisted coding ("vibe coding") requires two axes to evaluate: autonomy level (how much building is delegated to AI) and "operator discipline" (how much context persists between sessions as inspectable state rather than being reconstructed each ti...