Developer Tests Two Local AI Agents Building an App Autonomously Using Ollama

·5 views

A software developer ran an experiment using two locally hosted AI models — one acting as a code builder and one as a reviewer — to autonomously develop an application without human intervention. The setup, dubbed RelayLab, used Ollama and Python on a personal PC, with the models passing work back and forth across multiple rounds. Initial attempts with larger models caused CPU and VRAM overload, prompting a switch to smaller 3B and 4B parameter models with memory unloading enabled between turns. The orchestrator restricted agents to an isolated workspace, logging every round for post-experiment review. The experiment produced mixed but revealing results, highlighting both the potential and current limitations of local multi-agent AI systems.

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)

How unprepared management derailed a SaaS project and stalled a developer's career

A developer identified in the account as Maria spent seven months building a solid frontend from scratch for a SaaS overhaul at a small tech firm, only to have her work discarded when a newly hired senior developer convinced management to adopt a purchased template instead. The decision was made based on seniority rather than technical merit, setting a pattern where proven contributions were overridden without structured debate. When a core technology update caused the template to break four months before launch, neither the senior developer nor the team could resolve the incompatibility, forcing the project to be rebuilt on a third template. A project originally scoped for under one year stretched beyond two years with no clear completion date, accumulating technical debt at each restart. The account highlights how amateur project management — treating technology choices and deadlines as unrelated decisions — ultimately burdens frontline developers while shielding decision-makers from accountability.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Stored XSS Flaw Found in Next.js Demo Store via Malicious SVG Upload

A stored cross-site scripting (XSS) vulnerability was discovered in OopsSec Store, a deliberately vulnerable Next.js e-commerce app used for security training. An attacker with admin access could upload a malicious SVG file as a product image, embedding JavaScript that executes automatically for every visitor who views the product page. The flaw existed because the server trusted the client-supplied Content-Type header without inspecting actual file contents, and the app rendered SVGs using an img tag that allowed script execution. The vulnerability affected multiple endpoints, including the product page, admin preview, and the direct file URL. Recommended fixes include magic-byte file inspection, server-side SVG sanitization with DOMPurify, and strict Content-Security-Policy headers on the uploads directory.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

broot's whale-spotting mode and staging area streamline disk cleanup in one pass

When disks fill up, engineers typically rely on repetitive du and rm commands that only show flat directory totals, requiring deletions one folder at a time. The broot file manager offers an alternative workflow using its whale-spotting mode, launched with br -w, which sorts files and directories by size in an interactive tree. Users can press Ctrl-G on any file to add it to a staging panel, building a review set and seeing the total space to be freed before deleting anything. Once the desired files are collected, switching focus to the staging panel with Ctrl-Right and typing :rm removes the entire set in a single step. The workflow reduces disk cleanup to three stages — list, review, and delete — replacing the traditional back-and-forth loop with a single interactive session.

0 comments Read more at DEV Community

ProgrammingHacker News ·

Zelensky warns Russia expanding strikes to target Ukraine's data centres

Ukrainian President Volodymyr Zelensky has stated that Russia has broadened its military attacks to include strikes on Ukraine's data centres. The escalation marks a shift in targeting strategy, moving beyond traditional infrastructure to digital and technological facilities. Zelensky made the announcement as part of ongoing efforts to highlight the evolving nature of Russian aggression against Ukraine. The targeting of data centres raises concerns about disruption to critical digital services and information infrastructure within the country.

0 comments Read more at Hacker News