// edition · 2026-06-20
22 stories on AI dev tools, agents, and the coding stack, curated from the day's RSS haul by Agentic Dev's pipeline.
A developer spent $8,857.62 over 14 days using Claude Code's API to build six projects, consuming 3.884 billion tokens across 47,235 requests on Claude Opus 4.8. Cache hits accounted for 86.4% of total tokens, with the largest single project costing approximately $4,200.
A developer built a multi-agent Claude Code workflow called `/wizard` that uses an orchestrator agent to delegate coding tasks to specialist subagents, each working in isolated git worktrees in parallel. The system adds structured practices—codebase review, acceptance criteria, and test-first dev...
A developer published a guide on Dev.to comparing multiple Claude models within Anthropic's Claude Code environment, recommending Sonnet 4.6 as a default workhorse at $3/$15 per million tokens and Opus 4.8 for architecture and judgment-heavy tasks.
GitHub built an internal analytics agent called Qubot, powered by GitHub Copilot, that allows employees to query the company's data warehouse in plain language via Slack, VS Code, or the Copilot CLI. The tool generates SQL queries against Trino and Kusto databases and returns answers within secon...
Anthropic's Claude API can be integrated into C# .NET applications using the official Anthropic.SDK NuGet package, which supports .NET 6+ and features including streaming, tool use, prompt caching, and multi-turn conversations. The package is installed via `dotnet add package Anthropic.SDK` and r...
A guide published by claudeguide.io describes how to track Claude API spending by logging input and output token counts from API responses and aggregating costs in a time-series database. An audit of 12 Claude API deployments cited in the guide found model routing errors in 9, averaging $180 per ...
Anthropic's Claude API uses keys generated at console.anthropic.com, stored as the `ANTHROPIC_API_KEY` environment variable, which the SDK reads automatically. Keys begin with `sk-ant-api03-` and are displayed only once upon creation.
Retrieval-Augmented Generation (RAG) is a technique for grounding large language model outputs in external documents by retrieving relevant content at query time, adding it to the prompt, and generating answers from that context rather than training data alone.
A tutorial published on Dev.to demonstrates building a zero-shot support ticket triage system using the llama-3.3-70b model via the Oxlo.ai API, classifying tickets into five categories and four urgency levels using only prompt instructions and no training examples.
Google released the Flutter GenUI SDK (version 0.9.2) at Google I/O 2026 alongside Flutter 3.44, implementing the A2UI v0.9 protocol. The SDK allows Gemini AI models to generate Flutter app interfaces at runtime by parsing structured JSON into widgets, with developers supplying a predefined widge...
Checkmarx unveiled a new static application security testing engine combining a deterministic rules-based scanner, a fine-tuned LLM, and a false-positive classification layer called the Findings Analysis Engine. The company reported an F1 score of 0.499 against a category average of 0.20, and sai...
At the Observability Summit in Minneapolis, attendees discussed a shift from reactive to proactive monitoring, with sessions focusing on AI-driven incident response and noise reduction. Hud CTO May Walter presented on agentic issue resolution, arguing that production runtime context should feed i...
Headroom is an open-source context compression layer for AI coding agents that intercepts tool outputs, logs, and files before they reach an LLM, reducing token usage by 47–92% in tested workloads. It is available as a Python/TypeScript library, a local proxy, or an MCP server, and claims to pres...
A developer configured Claude's MCP feature to connect Notion and Google Docs in read-only mode, allowing Claude to reference project documentation across separate sessions without requiring manual context re-entry each time.
PDFMakerAPI released an MCP server that lets AI agents generate editable PDFs via a single `create_document` tool, configurable in clients like Claude Desktop or Cursor with a short JSON snippet. The server requires no API key and is also available as a hosted endpoint at api.pdfmakerapi.com/mcp.
Avital Tamir, a software engineer at cloud observability firm groundcover, argues that AI code reviewers are more reliable than human reviewers for catching routine errors, citing slow turnaround times and low-context peer reviews as evidence that traditional code review processes need revision.
Anthropic updated Claude Design with a bidirectional integration with Claude Code, allowing designers and developers to sync work via `/design-sync` and `/design` terminal commands. The update also added brand consistency features and connectors to tools including Adobe, Canva, Vercel, and Figma,...
A developer released "The Deal Team," an MIT-licensed set of 20 Claude-based AI agents for M&A due diligence that requires every factual claim to cite a source document or be discarded before reaching the user. The system covers five deal stages and is available on GitHub.
WAIaaS is an open-source, self-hosted Wallet-as-a-Service system designed to let AI agents execute payments autonomously within administrator-defined policy limits. It uses a three-tier authentication model and supports HTTP-level payment protocols, cross-chain bridging, and DeFi integrations.
Anthropic's Claude Design, launched April 17, 2026, generates editable visual outputs — prototypes, slide decks, and mockups — rendered as HTML/CSS, with export options to Canva, Adobe, Figma, and PowerPoint. The tool reached one million users within a week of a June 2026 update that added design...
Sean Lynch, in a Hacker News comment, argued that MCP's primary value over skills or CLI tools is isolating authentication flows outside an agent's context window, suggesting its ideal form may be an auth gateway for APIs.
The US government ordered Anthropic to withdraw its Fable 5 and Mythos 5 models, citing national security concerns after Amazon researchers identified a method to bypass Fable 5's safety guardrails. Cybersecurity researchers signed an open letter opposing the move, while Anthropic noted the same ...