GitLab introduces Duo Agent Flows to bring pipeline-level controls to AI agents

·1 views

GitLab this week announced Duo Agent Flows, a new abstraction that packages multi-step AI agent runs as named, managed units within its platform. The flows are designed to carry the same access controls, triggers, and review gates that apply to standard CI/CD pipeline objects. GitLab highlights three example use cases: implementing an issue, fixing a pipeline, and reviewing a merge request. A key claim is policy inheritance — when GitLab updates a permission or review gate, existing flows automatically reflect the change without manual edits across repositories. However, the announcement does not yet detail per-flow cost caps, execution isolation, rollback mechanisms, or how misbehaving flows will be handled at scale.

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)

Developer Builds Free Browser-Based Astigmatism Screening Tool Using Siemens Star

A developer has created a free online astigmatism screening tool at sighthealthpro.com, designed to take under 30 seconds and require no account or signup. The tool uses a Siemens Star — a radial pattern of black and white lines — where blurry or grey lines in certain directions can indicate astigmatism. Astigmatism affects roughly one in three adults, many of whom are unaware they have the condition. The tool is intended as a screening aid, not a medical diagnosis, and users flagged by the test are advised to consult an optometrist for a proper examination.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

10 Questions to Evaluate AI Memory OSS as a Backend System, Not Just Prompts

A developer guide published on DEV Community proposes 10 structured questions for evaluating AI memory open-source software — such as Honcho — as production backend systems rather than simple prompt utilities. The framework covers the full memory lifecycle, from raw message storage and asynchronous LLM processing to vector index synchronization, tenant-scoped retrieval, and deletion handling. Key concerns include how derived AI-generated claims are separated from original user messages, how idempotency and worker failures are managed in queue-based pipelines, and how mismatches between the database and vector index are detected and reconciled. The guide uses Honcho's pinned source code and changelog as a case study, noting features like work_unit_key for deduplication and a reconciler for sync-state recovery. The author emphasizes that operational reliability depends on explicit backend contracts — covering provenance, failure boundaries, and schema evolution — rather than on model output quality alone.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Full Stack Developer Krishnan Launches AI-Powered Interactive Portfolio Site

Krishnan, a full stack developer, has launched a personal portfolio website at krishnan-showcase.netlify.app showcasing his web development work. The site is built using React, Three.js, Tailwind CSS, and Netlify Functions, and features modern animations along with an AI-powered portfolio assistant. It highlights his featured projects, technical skills, and interactive UI/UX experiments. Krishnan specializes in building fast, responsive, and SEO-friendly websites as well as modern SaaS products. He is currently open to opportunities and seeking to connect with developers, recruiters, and potential clients in the frontend and full-stack space.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Website or Mobile App First? How to Make the Right Call for Your Business

Choosing between a website and a mobile app is a critical early decision for businesses launching digital products in 2026. Websites are generally recommended as a starting point for most startups due to faster development, lower costs, easier updates, and better search engine discoverability. Mobile apps become the stronger choice when a product requires frequent daily use, device-specific features like camera access, or highly personalized user experiences with push notifications. Experts suggest a phased approach: launch a website first, gather real user data, and then build an app only if it would meaningfully improve the experience. This strategy helps businesses avoid unnecessary spending and ensures product decisions are driven by actual customer behavior.

0 comments Read more at DEV Community