Engineer Runs AI Agent on Live Cloud Ops for a Week, Documents Wins and Failures

·1 views

A software engineer spent seven days testing an AI agent on a real cloud environment, granting it read access and limited write permissions to handle routine operations tasks. The agent handled triage faster than manual methods, correctly diagnosing 8 of 11 alerts and cutting mean resolution time from roughly 45 minutes to 20. However, it was confidently wrong three times, once misattributing an outage to a recent deployment when a downstream database was actually at fault. In one near-miss, the agent proposed scaling a broken service from 3 to 30 replicas, an action that was blocked by a mandatory human-approval guardrail. The engineer concluded that AI agents add real value in routine triage and context-gathering, but human oversight remains essential for any action that changes production state.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Vite 8 Launches with Rust-Based Rolldown, Cutting Build Times Up to 30x

Vite 8 was released in March 2026, introducing Rolldown, a Rust-based bundler developed by the VoidZero team that replaces both esbuild and Rollup in the production pipeline. The upgrade delivers 10–30x faster build times on real-world projects, with Linear reporting its build time dropping from 46 seconds to 6, and GitLab seeing a 7x reduction. Alongside Rolldown, the Oxc toolchain now handles TypeScript stripping, JSX transforms, minification, and tree-shaking, eliminating the need for Babel in React projects and reducing dependency size from roughly 45MB to 8MB. Vite 8 also introduces native TypeScript path alias resolution and an optional Full Bundle Mode that extends bundling to the development server for faster startup on large projects. The release requires Node.js 20 or higher, and most existing Vite 6 or 7 projects can upgrade with little to no configuration changes due to a built-in compatibility layer.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Why OAuth Consent Screens Matter More Than Users Realize

OAuth 2.0 consent dialogs — the permission prompts users see when linking third-party apps — are critical trust moments that most people dismiss in seconds without reading. The core mechanism behind these screens is scopes, which define what access an application is requesting, but scopes are often too broad to give users a clear picture of what they are actually agreeing to. Extensions like Rich Authorization Requests and Pushed Authorization Requests have been standardized to address cases where basic scopes fall short, enabling more precise and context-aware permission descriptions. Best practices recommend requesting granular, directional scopes and presenting permission prompts at the moment a relevant feature is used, rather than bundling everything at initial signup. Since consent grants typically persist until manually revoked, a vague or poorly timed consent screen can leave users unknowingly exposed to broad data access for months or years.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

DZHC: Netherlands-Based Firm Claims to Operate Entirely Without Human Employees

Dutch Zero-Human Company (DZHC), incorporated in the Netherlands and operational since early 2026, runs its engineering, research, finance, and client communications entirely through AI agents. The company uses Anthropic's Claude models coordinated via an orchestration layer called Paperclip, with humans involved only at the board level for strategy and high-stakes approvals. DZHC describes its model as a 'Zero-Human Company' and says it aims to demonstrate that AI-governed operations can work in practice, not just in theory. Alongside running itself autonomously, DZHC sells infrastructure tools to other organisations building similar setups, including decentralised identity credentials, on-chain treasury management via Safe multisig on Base mainnet, and governance design frameworks. The company says it publishes its governance policies and agent identities publicly to provide the broader field with real operational data rather than marketing claims.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Somali Developer Launches AI Chatbot and Courses Platform Called Isbar-Si AI

A developer named Faratoon has released Isbar-Si AI, an open-source web platform designed to deliver AI-powered courses and a Somali-language chatbot. The platform offers four main courses covering topics such as AI, video editing, and dropshipping, alongside three private classes in mathematics, programming, and the Canadian curriculum. Users can also book online appointments and access a structured knowledge base of reference articles. The site features an AI popup chatbot that automatically opens four seconds after the page loads and communicates in Somali. Built with React 19, Node.js, TiDB, and a built-in LLM, the project is publicly available on GitHub for self-hosting on a Ubuntu VPS.

0 comments Read more at DEV Community