Why I chose LangGraph over LangChain for multi-agent orchestration
State machines beat chains when your agents need to loop, retry, and share context. Here is the architecture that drove the decision.
Now in development - shipping Q3 2026
Your AI
senior reviewer
never sleeps
BugLens reviews every pull request before your team does - catching bugs, vulnerabilities, and style violations using your own codebase as context.
$ buglens review --pr 142
Fetching diff... 847 lines changed across 12 files
Running Lens agent -> Context agent -> Review agent
[!] auth/middleware.ts:34 - SQL injection risk
// User input directly interpolated into query string
[x] api/upload.ts:89 - No file size validation
// Max 10MB enforced in docs but not in code (see RFC-22)
[ok] utils/cache.ts - Matches team caching standard
-> Posted 3 inline comments to PR #142
-> Severity score: 6.4 / 10 | Request changes: yes
$
// how it works
Three agents.
One verdict.
BugLens uses a LangGraph pipeline of three specialised AI agents that work in sequence on every PR diff.
Lens agent
Parses diffs with AST analysis. Detects OWASP vulnerabilities, logic errors, and anti-patterns at the line level.
Context agent
Searches your team's docs, past PR comments, and standards via RAG. Every review knows your codebase.
Review agent
Writes structured inline comments with severity scores and one-click suggested fixes - posted directly on the PR.
GitHub native
Install via GitHub App in 60 seconds. Works on any repo, any language. No config files needed to get started.
MCP server
Expose BugLens agents to any MCP-compatible AI tool. Let your IDE assistant query review history and standards.
Review analytics
Track bug patterns, recurring violations, and team-wide code health over time on your BugLens dashboard.
// from the builder's log
Building in public
Technical deep-dives on AI agents, RAG pipelines, and the engineering decisions behind BugLens.
How BugLens uses RAG to review code with your team's own standards
Generic LLM reviews are useless without context. Here is how we embed docs, past PRs, and RFCs into every review.
AST-based SQL injection detection: lessons from EnginIQ
Before BugLens, I built EnginIQ to protect Postgres from AI-generated SQL. The same AST techniques now power BugLens's scanner agent.
MCP in 2026: why it is the USB-C moment for AI agents
Model Context Protocol went from niche spec to the default integration layer for AI tooling. Here is what that changes.
Follow the build
New post every week. No spam - just honest engineering notes from building BugLens in public.