Hey, I'm Andrew Larsen

I'm a software developer, consultant, entrepreneur, CTO, & co-founder of Compoze Labs.

I've been building software for 15 years, building teams for 8, and running an engineering organization for 5.

I started building software in the JVM world, and will always have a soft spot for Java. I quickly moved up the stack into the Cloud world, AWS will always be my favorite, but I have built and designed a lot of workloads in Azure and GCP. I'm obsessed with developer experience, and in 2025 I became officially AI-pilled.

You can find my writing and videos that have influenced my view of the software world as well as work that I think is worth sharing.

GitHub LinkedIn Medium

CONTENT

Things I've published

Practical Cloud Architecture

Cloud architecture isn't about using every service available—it's about choosing the right tools for the job and optimizing for both performance and cost. I focus on practical, real-world patterns that balance scalability with maintainability.

How to reduce your AWS NAT Gateway bill

Cost optimization strategies for AWS networking infrastructure, covering NAT Gateway alternatives and best practices for reducing cloud networking expenses.

AWS NAT Gateway, Fargate, & ECR Cost Gotcha

A deep dive into the hidden costs of AWS NAT Gateway when using Fargate and ECR, and how to avoid unexpected networking charges.

AWS VPC Network Security: Segmenting Your Network

Best practices for securing your AWS infrastructure through proper VPC network segmentation and isolation strategies.

Publish AWS CloudWatch Metrics with Node.js and TypeScript

A practical guide to publishing custom CloudWatch metrics from your Node.js applications using TypeScript and the AWS SDK.

AWS Certificate Manager DNS Validation with Terraform

Automating SSL/TLS certificate provisioning and DNS validation using AWS Certificate Manager and Terraform.

Cleanup S3 buckets with Go & Cobra CLI

Practical guide to building a CLI tool for automated S3 bucket cleanup during infrastructure teardown using Go and Cobra.

Adding Custom Cognito SignUp Message with Terraform and Lambda

Implementing custom email confirmations for AWS Cognito user signups using Lambda triggers and Terraform configuration.

Secure By Default: Using HashiCorp Packer With AWS and EBS

Creating Amazon Machine Images with encrypted root devices using Packer—a repeatable solution for meeting PCI, HIPAA, and FISMA compliance requirements.

RECOMMENDATIONS

Content I Love

These are the articles, talks, and resources that have shaped how I think about software development. Each one changed my perspective or clarified something I'd been struggling with.

Video

TDD, Where Did It All Go Wrong (Ian Cooper)

This talk fundamentally changed how I think about TDD. Ian Cooper clarifies what Kent Beck actually meant—test behaviors, not implementation details. It's the antidote to the brittle test suites that make people hate TDD. Every developer should watch this.

Article

Mocks Aren't Stubs

Martin Fowler's classic article that finally made the mock vs stub distinction click for me. Understanding when to use each approach has made my tests clearer and more maintainable. This is essential reading for anyone serious about testing.

Article

Three Pillars

Fowler's reflection on Thoughtworks' three-pillar model—Sustainable Business, Software Excellence, and Social Justice—resonates deeply with how I think about building companies. It's a reminder that great software companies balance commercial success with craft and social responsibility.

Article

Code Review Essentials for Software Teams

Blake Smith nails what code review is actually for—keeping the team mentally aligned, not nitpicking style. His hierarchy of needs for code review (team alignment > correctness > design discussion > style) has shaped how I approach PRs. The advice on framing feedback as questions is gold.

Article

Semantic Diffusion

Fowler explains how useful terms lose their meaning as they spread—'agile' becoming meaningless, 'refactoring' used for any code change. Understanding this phenomenon helps you recognize when buzzwords have drifted from their original intent and why going back to primary sources matters.

Video

No Vibes Allowed: Solving Hard Problems in Complex Codebases

Dex Horthy's talk on context engineering for AI coding tools is essential viewing. His 'frequent intentional compaction' approach and the concept of 'dumb zones' vs 'smart zones' has directly influenced how I use AI tools in production codebases. Practical, measured, and immediately applicable.

Tools & Practices

AI-Augmented Development

I've spent the past year working intensively with coding agents—Claude Code, Cursor, and others. These are my hard-won convictions about what actually works.

My Philosophy

These principles naturally lead to workflows like Research → Plan → Implement and patterns like Spec-Driven Development. The specific terminology will suffer from semantic diffusion eventually, but the underlying practices are sound.

Context is King

LLM performance degrades significantly as context windows grow. This isn't a minor detail—it fundamentally shapes how you should work. Keep sessions focused. Know when to start fresh and pass context forward. Use compaction strategies (like Claude's /compact) before you hit the wall, not after.

Planning First, Always

Stop trying to one-shot features. The pattern that consistently works: start with "help me figure out how to build this," flesh out product and technical details, and only move to implementation when the plan is solid. Most harnesses have planning modes now—use them. Track plans in markdown files. Your future self (and your future agents) will thank you.

Version Your Decisions

A lot of valuable context evaporates when we don't write down the decisions we make with our agents. I think of this like the ADR pattern, but for both human and AI consumption. Don't track everything—track the important decisions. Plans, architectural choices, the "why" behind non-obvious implementations.

Humans Still Matter

Based on research, trusted industry voices, and my own experience: a human still needs to be in the loop. Not as a rubber stamp, but actively steering technical direction, product direction, and maintaining the context needed to support the software long-term. Think of it as code review—but for the entire development process.

Claude Code Configuration

The ecosystem is standardizing around a set of tools and configurations. Select a configuration to see an example.

CLAUDE.md

Project context and instructions loaded at startup

~/.claude/ (global) or ./CLAUDE.md (project)

Example

# Project: My Application

## Overview
This is a Next.js application using TypeScript, Tailwind CSS, and shadcn/ui.

## Architecture
- /app - Next.js App Router pages
- /components - Reusable React components
- /lib - Utility functions and configurations

## Conventions
- Use functional components with TypeScript
- Prefer server components where possible
- Use Tailwind CSS for styling
- Follow the existing patterns in the codebase

## Important Notes
- Always run tests before committing
- Use conventional commit messages
- Keep components small and focused

Commands & Workflows

A collection of slash commands and workflows I use regularly. Select a command to view its documentation or watch a demo.

/feature:research

Feature Development

Document codebase using parallel Explore agents. Spawns locator, analyzer, and pattern finder agents to map code without suggesting improvements. Outputs to docs/research/.

Usage

Document codebase using parallel Explore agents. Spawns locator, analyzer, and pattern finder agents to map code without suggesting improvements. Outputs to docs/research/.

# Invoke in Claude Code:

/feature:research

View full source at: .claude/commands/feature/research.md