// edition · 2026-05-13
38 stories on AI dev tools, agents, and the coding stack — curated from the day's RSS haul by Agentic Dev's pipeline.
Claude Code's Routines feature lets users configure automated cloud-based jobs that run on schedules, API triggers, or GitHub events without requiring a local machine. Usage limits are 5 routines per day on Pro, 15 on Max, and 25 on Team and Enterprise plans.
GitHub updated its Copilot individual plans ahead of a June 1 shift to usage-based billing, adding a "flex allotment" that boosts included usage: Pro ($10/month) gets $15 in total usage, Pro+ ($39/month) gets $70, and a new Max plan at $100/month includes $200 in total usage.
A developer published a practical guide to Anthropic's Claude SDK Tool Use feature, using version 0.101.0, covering how to build agents that delegate tasks like date arithmetic and API lookups to external functions rather than relying on the model to compute answers directly.
A developer guide describes using Claude's tool_use API feature to guarantee schema-compliant JSON output by defining a fake tool with the desired JSON Schema, forcing Claude to return structured data as tool call arguments rather than free-form text.
A developer found that Cursor sent 8,400 input tokens for a simple function rename request, compared to 1,900 tokens when making the same call directly to the Anthropic API. Facing 50% month-over-month growth in API costs, the developer cancelled Cursor and built a 200-line replacement, reducing ...
Claude Code can be configured to route requests through Microsoft Azure AI Foundry by setting environment variables and authenticating via Azure CLI, allowing users whose employers already have Claude models deployed on Foundry to avoid a $17/month personal subscription fee.
Anthropic's Claude Code includes a memory consolidation feature called Dreaming, which reads project memory files, removes stale entries, and merges duplicates into a condensed version. It runs automatically after roughly 24 hours and five sessions of inactivity, or can be triggered manually with...
NVIDIA engineers and researchers use OpenAI's Codex with GPT-5.5 to build production systems and conduct research experiments. The collaboration covers both software development and experimental research workflows.
Vercel added a fast mode for Claude Opus 4.7 on its AI Gateway, offering approximately 2.5x faster output token generation in research preview. The feature is priced at 6x standard Opus 4.7 rates, with input at $30 per million tokens and output at $150 per million tokens.
Organizations deploying AI coding agents in regulated CI/CD environments are encountering compliance gaps because agent-initiated changes lack auditable records of inputs, prompts, policy checks, and decision chains. A financial institution case illustrates the problem: when auditors requested pr...
A Dev.to guide outlines an AI-assisted MVP development workflow using tools including Cursor, v0 by Vercel, Supabase, Replit Agent, and Claude API, claiming founders can ship functional prototypes in two to four weeks versus the traditional three to six months.
A developer tutorial describes how to integrate the x711 API with the OpenAI Agents SDK using a single HTTP endpoint to expose 26 tools, including web search and price feeds, as Python functions with docstrings.
A developer found that using Claude's Plan mode before writing code, combined with writing tests during implementation, reduced token waste and produced more consistent results than jumping straight into implementation.
A developer published a guide describing how teams can share Claude AI prompts, skills, and commands by hosting a Git repository structured as a plugin marketplace, using a `marketplace.json` index file compatible with both Claude Code and Cowork clients.
Cybersecurity researchers are warning that enterprise AI agents, which have broad access to company data and systems, introduce new attack vectors where malicious actors can exploit agents' instruction-following behavior to exfiltrate sensitive information, a tactic being called "living off the a...
Vercel added CLI support for its firewall product, allowing users to configure custom rules, IP blocks, system bypasses, attack mode, and DDoS mitigations via the `vercel firewall` command. A companion Vercel Firewall skill enables AI agents to interact with firewall settings.
SAP announced a fully managed version of Joule Studio at its Sapphire 2026 conference, adding support for Cursor and Claude Code alongside agent frameworks AutoGen and LlamaIndex. The platform includes a new SAP Domain Models family and offers 12 months of free design-time access, with general av...
x711 is a service that provides a single API endpoint and key to supply CrewAI agents with tools such as web search and price feeds, replacing the need for multiple separate API integrations. Users obtain a key via a POST request to x711.io and route tool calls through one endpoint.
OpenAI published a guide describing how finance teams can use its Codex coding assistant to automate tasks including monthly business reviews, reporting packs, variance bridges, model checks, and planning scenarios.
A developer built "GitHub Dungeons," a GitHub CLI extension written in Go that converts a code repository into a playable terminal roguelike game. Dungeon layouts are procedurally generated using Binary Space Partitioning, seeded by the repository's latest commit SHA, so each commit produces a di...
At Google Cloud Next 2026, Finout CEO Roi Ravhon and Google Cloud FinOps lead Pathik Sharma said enterprises face rising AI costs despite falling token prices, because newer reasoning models consume more tokens per task and token usage is variable for identical prompts. Both recommended routing A...
Researchers developed a reinforcement learning method to train language models to self-correct their own outputs, addressing a limitation where models struggle to identify and fix their own errors without external feedback.
Simon Willison built a browser tool that intercepts CSP violations in sandboxed iframes and prompts users to add blocked domains to a fetch() allow-list, which then refreshes the page. The tool was developed using GPT-4.5 in the Codex desktop app.
Red Hat announced Red Hat AI 3.4 at its Summit in Atlanta, adding Model-as-a-Service capabilities that provide a shared API interface for accessing pre-trained models with usage tracking and policy enforcement. The release also includes request prioritization for distributed inference and specula...
Vercel added Node.js 26 support to its Sandbox environment. Users can access it by upgrading the @vercel/sandbox package to version 1.10.2 or later and setting the runtime property to "node26".
SerpApi provides an API that returns structured JSON from search engines including Google and Amazon, handling proxy rotation, CAPTCHA bypassing, and parser maintenance on behalf of developers. The service targets AI teams that would otherwise build and maintain custom web scrapers to access live...
SimplyLouie, an independent service, offers API access to Anthropic's Claude AI model at ZAR 37/month, targeting South African developers priced out of ChatGPT Plus and GitHub Copilot, which cost approximately ZAR 370/month. The company states half its revenue is directed to animal rescue operati...
Vercel introduced "Trusted Sources," a deployment protection method that accepts short-lived OIDC tokens from authorized Vercel projects and external services, replacing long-lived automation bypass secrets. Callers pass tokens via the `x-vercel-trusted-oidc-idp-token` header; Vercel verifies the...
Nvidia released OpenShell, an Apache 2.0 open source secure runtime for autonomous AI agents, built over six months by senior director Ali Golshan's team. The system isolates each agent in a sandbox with an external gateway handling credentials, preventing agents from directly accessing host infr...
API evangelist Kin Lane argues that organizations with mature API management and well-maintained OpenAPI specifications are better positioned to adopt AI agents, noting that the Model Context Protocol (MCP) is a long-lived HTTP connection serving JSON that can be generated directly from existing ...
x711.io offers an MCP server at x711.io/mcp that provides 29 tools — including web search, cryptocurrency price feeds, transaction simulation, and code execution — accessible in Claude Desktop, Cursor, Cline, or Windsurf via a single JSON configuration entry. Web search is free up to 10 queries p...
Simon Willison released llm 0.32a2, an alpha version of his LLM command-line tool. The update switches reasoning-capable OpenAI models to the `/v1/responses` API endpoint, allowing summarized reasoning tokens to be displayed during prompts, with a `--hide-reasoning` flag to suppress them.
Frederick P. Brooks' 1986 paper "No Silver Bullet" argued no single technology would dramatically improve software developer productivity. The Pragmatic Engineer reexamines that thesis in light of AI coding tools and agents that now generate substantial amounts of code.
Industry surveys show 73% of engineering organizations reduced junior hiring over two years as AI tools help developers complete tasks up to 55% faster, while JetBrains' 2026 data puts AI coding assistant adoption at 18% globally. Critics argue the speed gains mask a skills gap, as junior develop...
A software developer argues that AI code generation tools remain limited in handling ambiguous requirements, security vulnerabilities, and long-term maintainability, citing figures such as 45% of AI-generated code samples containing vulnerabilities and AI accruing technical debt twice as fast as ...
Datawhale, a Chinese AI learning community, published Easy-Vibe, an open-source programming curriculum built around AI-assisted "vibe coding," accumulating over 10,300 GitHub stars. The three-stage course targets non-programmers and covers tools such as Cursor and Claude Code, with support for 10...
OpenAI's Parameter Golf competition drew over 1,000 participants and 2,000+ submissions focused on AI-assisted machine learning research, coding agents, quantization, and model design under strict parameter constraints.
SAP launched the AI Agent Hub at its Sapphire 2026 conference, a vendor-agnostic registry for managing AI agents, LLMs, and MCP servers across enterprise environments. Two of six planned features are generally available now, with the remaining four, including identity management and observability...