How a Layered AI Code Review Pipeline Catches Bugs Human Reviewers Miss

·1 views

A software developer has detailed a structured code review pipeline that combines deterministic tools with a large language model (LLM) to improve bug detection in pull requests. The pipeline runs formatters and linters first as blocking steps, reserving the LLM only for semantic issues that static tools cannot detect. This layered approach addresses a common failure mode where AI reviewers flood developers with noise — including style comments and false positives — causing teams to ignore them entirely. The LLM layer is specifically scoped to catch issues like swallowed exceptions, missing awaits, N+1 database queries, and changes that contradict a pull request's stated intent. As of mid-2026, the author notes that model API costs make this approach viable at just cents per pull request, but emphasizes that pipeline design is more critical than the choice of model.

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 Fully Functional Windows 98-Themed Portfolio Without AI or Frameworks

A developer has hand-coded a Windows 98-themed portfolio website using only vanilla JavaScript, HTML, and CSS, deliberately avoiding AI tools and modern frameworks. The project features a working applet system that stores mini-applications in the browser's localStorage, mimicking the experience of installing software on a 90s PC. Applets include a playable DOOM integration via WebGL, a Half-Life soundboard using the HTML5 Audio API, and a 3D maze screensaver built with Canvas. The creator argues that AI-generated code cannot authentically replicate the imperfections — such as jagged icons and inconsistent spacing — that defined the Windows 98 aesthetic. While the approach increases debugging complexity and faces storage constraints from localStorage's 5MB limit, it reflects a deliberate effort to preserve the DIY ethos of early internet culture.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How to Fix Nextcloud IP Detection and HSTS Headers Behind Traefik and Cloudflare

A developer documented two key configuration issues encountered while hardening Nextcloud 29 running in Docker on a Proxmox LXC behind Traefik v3 and a Cloudflare Tunnel. The first problem involved Nextcloud misidentifying all client IPs as Traefik's IP, breaking brute-force protection and rate limiting, which was resolved by setting trusted proxies and Cloudflare CIDRs via the occ command-line tool. The second issue was that Traefik's built-in HSTS middleware failed to send Strict-Transport-Security headers because Cloudflare terminates TLS before traffic reaches Traefik, causing it to treat connections as non-HTTPS. The fix required adding an explicit custom response header in Traefik's dynamic configuration rather than relying on the stsSeconds directive. Additional hardening steps included setting a maintenance window, defining a default phone region, and installing apps such as Calendar, Contacts, and Talk directly via occ instead of the web UI.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Terraform Data Sources: How to Deploy EC2 Without Hardcoding IDs

A Terraform tutorial published on DEV Community demonstrates how to deploy EC2 instances into an existing VPC and subnet using data sources instead of hardcoded resource IDs. The guide walks through referencing a pre-existing VPC, subnet, and the latest Amazon Linux 2 AMI dynamically via Terraform's data blocks. By filtering resources using name tags and owner metadata, the configuration avoids manual ID management and reduces human error. This approach promotes reusability, as shared network infrastructure managed by other teams can be referenced without being recreated. The method also ensures EC2 instances always launch with the most recent AMI, improving reliability and scalability across environments.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Cloudflare Launches Kitesurf, an AI-Agent Browser That Trades Speed for Efficiency

Cloudflare announced Kitesurf on August 6, 2026, a browser designed specifically for AI agents rather than human users, available in beta via Browser Run after just twelve weeks of development. Built on Cloudflare Workers using components of the Blitz rendering engine, it exposes the Chrome DevTools Protocol to support compatible automation tools. Benchmarks published by Cloudflare show Kitesurf uses significantly less CPU and memory than a warm Chromium pool for screenshot and HTML-extraction tasks, but runs roughly 1.7 to 1.8 times slower in wall-clock time for the same jobs. The company acknowledged Kitesurf is not suited for video, WebGL, certain bot-challenge handshakes, or long authenticated sessions requiring persistent state, with Chromium retained as a fallback for those cases. The project illustrates a deliberate product trade-off: optimising for the resources most relevant to AI agent workflows rather than attempting to outperform full-featured browsers across every dimension.

0 comments Read more at DEV Community