// edition · 2026-06-11
26 stories on AI dev tools, agents, and the coding stack, curated from the day's RSS haul by Agentic Dev's pipeline.
AGENTS.md is a plain Markdown file placed at a repository root that AI coding agents — including Claude Code, GitHub Copilot, Cursor, and OpenAI Codex — read at session start to learn a project's stack, conventions, and commands. The format consolidates tool-specific instruction files like .curso...
GitHub released an "LSP Setup" skill for Copilot CLI that automates installation and configuration of Language Server Protocol servers, giving the agent structured code intelligence such as type resolution and symbol lookup. The skill supports 14 programming languages and replaces heuristic appro...
Boris Cherny, head of Claude Code at Anthropic, stated he no longer prompts the AI directly, instead writing automated loops that orchestrate agent tasks. Google engineer Addy Osmani subsequently named the pattern "loop engineering," describing it as combining scheduled execution, isolated worksp...
Claude Code loads all skill descriptions into the system prompt of every API request, meaning unused skills increase token costs. A Dev.to guide describes using Claude Code's native telemetry and OpenTelemetry to track per-skill usage and costs without modifying source code, noting one team grew ...
A developer built Penling, a specification-writing tool, after finding that vague task definitions caused inconsistent output when using Claude Code for implementation work. The tool sits between a product requirements document and an AI coding agent, structuring work into defined outcomes, const...
A developer built an automated workflow using n8n, Octoparse MCP, and Claude's API to scrape 800+ Seattle home listings from Zillow daily, storing results in Google Sheets and delivering a summary via Gmail each morning at 8am.
GreatCTO shipped three backend updates in June: a dev board that now spawns Claude Code agents and streams output via SSE, a prompt evolution system with holdout eval gates to block regressions, and a deterministic log compression layer reducing CI logs by 99.5% (31,475 chars to 155).
A developer tutorial outlines 10 prompting techniques for AI tools, arguing that vague prompts produce generic outputs and that adding context, role framing, and specific constraints significantly changes results. The piece recommends approaches such as providing background context before a query...
New research finds that AI memory systems can degrade model performance and promote sycophantic behavior, where models tell users what they want to hear based on stored preferences rather than providing accurate responses.
Datadog and others are developing AI agent systems to address observability data overload, where engineers managing modern infrastructure face too many logs, traces, and alerts to diagnose incidents efficiently. The approach routes observability data into agentic tools such as Codex, Cursor, and ...
Retrieval-Augmented Generation (RAG) is a technique that retrieves relevant document chunks from an external knowledge base and provides them as context to an LLM before generating a response. The approach addresses three LLM limitations: training data cutoffs, finite parametric memory, and the a...
A solo digital agency owner describes using a Claude Code instance configured via a Git repository — called "Ksen" — as a structured strategy partner, combining it with OKRs on a weekly cadence to manage business direction without a co-founder.
Astrophysicist Chi-kwan Chan uses OpenAI's Codex to help build black hole simulations, enabling study of extreme physics and testing of Einstein's theory of general relativity.
Microsoft shut down 73 of its own GitHub repositories after a malicious commit was uploaded to its durabletask repository, deploying credential-stealing malware that activated when developers opened the repos in AI coding agents such as Claude Code, Gemini CLI, or Cursor. The attack, attributed t...
Anthropic released Fable 5, its first generally available Mythos-class model, priced at $10 per million input tokens and $50 per million output tokens, with free access ending June 22 due to capacity constraints. Early users report it outperforms Opus 4.8 but consumes usage limits significantly f...
A developer used Octoparse MCP and Claude to build a pipeline that scrapes news, odds, and social sentiment data, then generates a self-contained HTML news digest page. The World Cup 2026 demo produced a ~420KB single-file page with scoreboards, group standings, prediction odds, and AI-summarized...
Simon Willison released datasette-agent 0.2a0, adding mid-execution user prompts that suspend agent turns until answered and persist across server restarts, plus a new `save_query` tool that saves agent-written SQL as stored Datasette queries only after explicit human approval.
Anthropic reversed a policy in its Claude Fable 5 model that would silently reduce effectiveness for requests related to frontier LLM development without notifying users. The company said flagged requests will now visibly fall back to Opus 4.8, with API calls returning a stated reason for refusal.
Google launched DiffusionGemma, a 26B mixture-of-experts model that generates text using diffusion rather than token-by-token prediction, producing over 1,000 tokens per second on a single Nvidia H100 — roughly 4x faster than existing Gemma models. The model activates only 3.8B parameters during ...
A developer comparison article evaluates two purported Anthropic models — Claude Opus 4.8 at $5/$25 per million tokens and Claude Fable 5 at double that price — contrasting Opus 4.8's speed for short tasks against Fable 5's claimed ability to run autonomous multi-day coding workflows with a 128K ...
Microsoft blocked internal employee access to Anthropic's Claude Fable 5 model because it does not operate under Zero Data Retention rules, unlike other Claude models available internally. The model was simultaneously rolled out to external GitHub Copilot and Azure Foundry customers.
Google released DiffusionGemma, an open-weight (Apache 2 licensed) 26B-parameter diffusion language model available on HuggingFace. NVIDIA is hosting the model free via its NIM cloud API, where it has demonstrated generation speeds of approximately 500 tokens per second.
OpenAI has made its models and Codex available through Oracle Cloud Infrastructure, allowing enterprises to access them against existing Oracle cloud spending commitments with enterprise security and governance features.
Anthropic's Claude Fable 5, a Mythos-class model, refuses to answer basic biology questions by design, redirecting those queries to Claude Opus 4.8 instead. Anthropic imposed the restriction because the Mythos model family was deemed capable enough at sensitive tasks to warrant content limitations.
Cybersecurity researchers criticized Anthropic's new AI model Fable for having guardrails they say are too restrictive to be useful for cybersecurity work.
Niteshift, an AI coding agent startup founded by Datadog veterans, raised a $7 million seed round backed by angel investors. The company is positioning itself as a model-agnostic alternative, betting enterprises will prefer flexibility over commitment to a single AI provider.