Run LLM code review and QA on every pull request
A team added LLM-driven PR review, architecture tests and documentation to their pipeline; merged PRs tripled and issues fell, with agents filling gaps from related Jira tickets. ·
Done with a custom-built or unnamed agent
- What they did
- The team already had given/when/then acceptance criteria, linting, static analysis, automated tests, and QA and UAT rounds. About two months ago they added static and LLM-guided architecture tests, LLM-written product documentation, and LLM PR review. Agents sometimes filled gaps in the criteria by finding related Jira tickets. Developers still step in on the suggestions.
- How it went
- Merged PRs tripled, issues are falling, and time-to-merge is about a day, though the author doesn't credit LLMs for that last one. Downsides: QA catches planning-level mistakes, product knowledge fades, and PRs grow bigger.
- Worth knowing
- LLM review can be over-defensive, for example repeatedly asking for locks that current business rules make unnecessary, so a developer has to judge each suggestion.
Try it yourself with GitHub Copilot
Set up automated review on every pull request in my repo [repo name]. For each PR, have it check the diff for bugs, missing tests and style issues, compare it against the linked ticket in [Jira/issue tracker] to see if it covers the requirements, and post a summary comment. Start by showing me the workflow file you plan to add and wait for my approval before committing anything.
Want help getting this working? Join the Vibe Coding Academy ↗ Run by Alex Finn, who also runs this site.
Source: HN · Nov 12, 2026
Related use cases
Build a C compiler with a team of parallel agents
Anthropic's engineering write-up on tasking Opus 4.6 agent teams with building a C compiler.
Prototype product features as a non-coding PM
A founder used Cursor agents to find prototype gaps, draw ASCII wireframes, and plan and implement a feature with parallel agents in worktrees.
Build a first real project with Replit Agent
A student completed a real build with Replit Agent, spending about $65 in agent usage and finding the cost unexpected.
Run parallel local coding agent sessions
A developer runs up to 10 coding agent sessions at once, using Claude Code and local models on RTX 3090s with cached contexts for daily development work.
Add a filtered report page to a Razor Pages app
The poster used Copilot agent mode in Visual Studio 2022 to build a charges summary report page with date and status filters, add a menu option, and write the git commit steps.
Hand migrations and bug-fix tickets to a coding agent
Reviewers gave Devin migration, data engineering, bug fix and refactoring tickets on a Next.js monorepo and Python pipeline over three weeks; it did best on migrations and framework upgrades.
Get 5 like this every morning
The best things people got an AI agent to do, each with the prompt to try it.