// edition · 2026-05-02
24 stories on AI dev tools, agents, and the coding stack — curated from the day's RSS haul by Agentic Dev's pipeline.
A developer building a SaaS AI coding product identified three production failure modes with MCP servers: built-in tools silently overriding custom sandboxed MCP tools, and environment variables from parent processes leaking into MCP child processes as tokens. The post includes production code us...
Claude Code loses all session context on restart; a developer workflow using a MEMORY.md file in the project root provides a 200-line persistent index of evolving project state, such as recent migrations and active decisions, which Claude Code reads at session start.
Cursor released a public beta TypeScript SDK on April 29 that lets developers build agents on its orchestration harness, model-agnostic and deployable to cloud VMs. The company reported agent usage grew 15x over the past year, with agent users now outnumbering Tab autocomplete users by 2:1.
Anthropic updated its terms on April 4th, barring Claude Pro and Max subscribers from using subscription limits to power third-party agent frameworks such as OpenClaw, requiring pay-as-you-go API billing instead. The change can increase costs up to 10x, while a new Fast Mode tier charges $30 inpu...
A developer released Governor, an open-source Claude Code plugin that compresses memory files, filters build/test log output, and adds usage telemetry to reduce token consumption. Small local benchmarks reported approximately 55% reduction in output tokens and 96% of noisy pytest output blocked.
A developer released Scout, an MIT-licensed self-hosted research agent that uses Claude Code to convert GitHub Issues into cited research reports published to GitHub Pages. The system includes a pre-research "sharpening" step that clarifies ambiguous queries before dispatching parallel sub-agents...
Braintrust and Trainline held a workshop in London on deploying agentic AI applications in production, focusing on evaluation, observability, and testing practices beyond prompt engineering. The article outlines how production AI systems require both traditional software engineering discipline an...
A Dev.to comparison of Claude Code 3.5 and Tabnine 2026 tested 50 developers on three coding tasks, reporting Claude completed work 35% faster on average, while Tabnine had lower completion latency (90ms vs. 120ms) and a 12% lower initial error rate.
Incredibuild announced Islo, a cloud sandbox that provides each AI coding agent its own persistent, isolated environment with scoped credentials and policy controls. The product addresses security and operational issues that arise when agents run on developer laptops, where they inherit all user ...
Google's Gemini API free tier caps usage at 15 requests per minute and 1 million tokens per day, with 429 errors triggered by rate limits and 503 errors by server overload. A developer published a Rust error-handling pattern for these responses, distinguishing retryable errors (429, 503) from non...
A Dev.to tutorial demonstrates using Anthropic's Claude 3.5 LLM to debug Go 1.23 microservices, highlighting its 200K token context window and ability to identify race conditions from concurrent map writes in HTTP handlers.
A developer released App Architect, a Claude artifact that guides users through five structured phases — user flow, page map, system design, spec, and test planning — to produce a test-driven development prompt for use in AI coding tools. The tool runs on the user's own Claude account and require...
Uber depleted its entire 2026 AI budget within four months by spending it on Claude Code, Anthropic's AI coding tool. The rapid spending indicates unexpectedly high adoption or usage costs among Uber's engineering teams.
Developer Robert Nowell released "skill-tree," a tool that analyzes Claude Code session history against 11 collaboration behaviors from Anthropic's February 2026 AI Fluency Index study, scoring users and assigning one of seven archetype cards. It is available as a Claude Code plugin and as an npm...
Libelo, a park and nature discovery platform, built an AI conversational assistant using Azure AI Foundry routed through their own API rather than called directly from the mobile app, citing security, monitoring, and resilience concerns. The implementation uses Azure Entra External ID for authent...
A student developer built VoteGuide AI, a conversational assistant for Indian voters that uses the Groq API with LLaMA 3.3 70B to answer election questions, check voter eligibility, and run voting simulations. The app, built with React 18 and deployed on Vercel, supports Hindi and role-based resp...
AgentDM is an agent-to-agent messaging platform built on MCP that allows AI agents to communicate and collaborate autonomously. In a demonstration, two Claude-based agents handled a full software feature from specification to merged pull request without human input.
A developer tested four free LLM endpoints — Cerebras (Qwen 3 235B), Groq (Llama 4 Scout), OpenRouter (Ling-2.6 Flash), and Cloudflare (GPT-OSS 120B) — for production use in an AI website builder. Groq achieved ~500 tokens/second, Cerebras produced the most coherent output for complex prompts, Op...
Mistral AI released Mistral Medium 3.5 and added cloud execution to its Vibe coding agent, allowing agents to run tasks in isolated sandboxed environments independent of a developer's local machine. The company also added a "work mode" to its Le Chat interface for running longer, parallel tool-ca...
A developer tested three AI models (Claude, Codex, Gemini) by having each independently review the same BigQuery deduplication pipeline plan, finding that each model caught at least one critical bug the others missed, including a silent SQL scoping error, a midnight-boundary race condition, and u...
IBM released Bob, an agentic software development platform, after deploying it internally to over 80,000 developers since June 2025. Surveyed users reported an average 45% productivity increase, with specific teams citing 69–70% time savings on coding tasks, though IBM notes the figures are self-...
A developer opinion piece argues that effective AI prompting requires the same engineering skills as traditional coding — decomposition, typed inputs, testing, and structured constraints — and that experienced programmers hold an advantage over non-coders using LLMs.
Apple shipped Claude.md configuration files inside its Apple Support app, revealing the company uses Anthropic's Claude AI in its development workflow. The files, intended to provide instructions to the Claude AI assistant, were accidentally included in the distributed app package.
Replit CEO Amjad Masad said he would prefer not to sell the company, speaking at a TechCrunch event in San Francisco. His comments came amid reports that rival Cursor is in acquisition talks with SpaceX at a reported $60 billion valuation.