// edition · 2026-05-23
26 stories on AI dev tools, agents, and the coding stack — curated from the day's RSS haul by Agentic Dev's pipeline.
Anthropic will retire the Claude Opus 4 and Sonnet 4 models (snapshots dated 20250514) on June 15, 2026, following their deprecation on April 14, 2026. The versioned aliases `claude-opus-4-0` and `claude-sonnet-4-0` will also stop working on that date; Claude 4.1 remains active until at least Aug...
Security researcher Joernchen discovered a remote code execution vulnerability in Claude Code version 2.1.118, which Anthropic has since patched. The flaw stemmed from a parsing anti-pattern that the author argues is common across AI developer tools.
A Dev.to article outlines an AI development stack for 2026 centered on Cursor, Claude, and n8n, arguing that versioned rule files and automated workflows deliver more productivity gains than raw model improvements, citing teams with rule libraries as 3x faster than those using ad-hoc prompts.
Current AI coding models such as Claude 4.x and GPT-4o interpret prompts literally rather than inferring intent, requiring developers to supply stack context, precise task descriptions, constraints, and expected output format. Structured prompts using XML tags show 30–39% improvement in response ...
A developer cancelled a $20/month Claude Cowork subscription after switching to OpenWork, a free open-source AI agent client that offers built-in OAuth-based MCP server connections and several free models including DeepSeek and Qwen3 without requiring a paid plan.
A developer analyzed Twitter's open-source recommendation algorithm repositories and built a Claude Code sub-agent to score tweet drafts pre-posting. The algorithm's heaviest positive signal is author-engaged replies (+75), while mutes or blocks carry a -74 penalty, nearly canceling 150 likes.
Virgin Atlantic used OpenAI's Codex to ship a revamped mobile app on a fixed holiday travel deadline, achieving near-total unit test coverage and zero P1 defects in the release.
Composio is a tool-calling platform that connects LLMs such as Claude directly to external services including GitHub, Slack, and Gmail, enabling automated task execution without manual copy-paste steps. The approach trades human oversight for throughput, with manual workflows remaining preferable...
OpenHuman is a context persistence system for AI tools that automatically harvests, compresses, and re-injects user activity data — including prompts, files, and workflow patterns — into future AI sessions. Its internal pipeline, nicknamed "TokenJuice," runs on a 20-minute cron job to maintain sy...
A PagerDuty survey found 84% of companies have experienced at least one AI-related outage, while 68% lose more than $300,000 per hour during system failures. The report identifies accumulated technical, automation, and integration debt as primary risks as AI deployments move from pilot to product...
Anthropic released Claude 4 on May 23, 2025, including Claude Opus 4 and Claude Sonnet 4, at its first developer conference. Claude Opus 4 scored 72.5% on SWE-bench, supports up to roughly seven hours of continuous task execution, and is priced the same as its predecessor; GitHub plans to use Cla...
Anthropic held its first developer conference, Code with Claude, in San Francisco on May 6, 2026, announcing a Managed Agents platform at $0.08 per session-hour and a "Dreaming" feature that consolidates agent memory between sessions. Legal-AI firm Harvey reported a roughly 6x rise in task comple...
Sub2API is an open-source API gateway written in Go that aggregates multiple Claude, OpenAI, and Gemini subscription accounts into a single unified API endpoint. The project, maintained by Wei-Shaw on GitHub under LGPL-3.0, includes multi-account management, token-level billing, and load balancin...
A developer at AgenticBoxes.email built and deployed an MCP server within 12 hours of a customer request, using Claude as an engineering assistant. The release used an opt-in beta model where AI agents received broadcast announcements and self-enrolled via API before the feature was made generall...
A developer tool called skill-tree analyzes Claude Code session history, scoring users across 11 AI collaboration behaviors drawn from a taxonomy based on 9,830 Claude conversations, then assigns one of seven archetype profiles and suggests unused behaviors to practice.
Developer and educator Josh W. Comeau argues that AI tools amplify the output of people with existing technical skills rather than serving as a substitute for them, with stronger foundational knowledge producing proportionally greater gains.
JetBrains VP Mikhail Vink claimed at Google Cloud Next that JetBrains is the only independent AI coding tool vendor, contrasting with Cursor's ties to xAI, Windsurf's acquisition split between Google and Cognition, and Copilot's Microsoft-OpenAI alignment. The company, which has 16 million users ...
Anthropic's Claude for Small Business, which includes native connectors to QuickBooks, HubSpot, Canva, and Google Workspace, identified 17 of 20 deliberately planted errors in a fictional nine-tab profit-and-loss spreadsheet in a structured test by The New Stack.
Gartner named OpenAI a Leader in its 2026 Magic Quadrant for Enterprise AI Coding Agents, citing the company's Codex product for innovation and enterprise-scale deployment.
Anthropic and OpenAI each launched enterprise AI deployment subsidiaries within 72 hours in May 2025. Anthropic's firm, backed by Blackstone and Goldman Sachs among others, targets mid-sized enterprises, while OpenAI's "DeployCo" acquired consulting firm Tomoro and its ~150 engineers with over $4...
AiFinPay released an open-source Python SDK for processing payments between humans and AI agents, available via pip as "aifinpay-agent" with source code published on GitHub.
A Dev.to opinion piece argues that by 2026, autonomous AI agents capable of handling software architecture, coding, and deployment are replacing single-purpose coding assistants, with parallel agent systems dividing roles across backend, frontend, and DevOps tasks.
AiFinPay released an SDK for integrating payment processing and payouts into AI agent workflows, available via pip as `aifinpay-agent`. The project is hosted on GitHub at github.com/AiFinPay/sdk.
AiFinPay released a Python SDK for enabling payment processing in AI agent applications, available via pip as "aifinpay-agent" with source code published on GitHub.
A hacker group called TeamPCP breached GitHub via a malicious VSCode extension installed by a developer, compromising at least 3,800 of GitHub's internal code repositories. The group advertised the stolen source code for sale on BreachForums; GitHub stated the affected repositories contained its ...
AiFinPay released an open-source Python SDK (`pip install aifinpay-agent`) designed to integrate payment processing functionality for AI agents into applications. The source code is available on GitHub at github.com/AiFinPay/sdk.