Building with AI, in the open.

Sam Keen

I'm Sam Keen. I work hands-on with AI-assisted development, then write and teach what actually holds up, through a weekly newsletter and live workshops. Author of Clean Architecture with Python, and formerly a generative AI architect at AWS.

25+ years shipping production software.

Formerly at
Amazon Web Serviceslululemon athleticaNike

Co-founder of Portland AI Engineers (1,300+ members)

The newsletter

Every week, I dig into what's actually working in the new AI abstraction layer so you can make informed decisions without drowning in hype.

Monday AI Review

Signal filtering: what matters this week in AI for developers.

Deep Dives & Insights

Tech Explorations & Industry Analysis. Go deeper than the headlines.

What makes it different

Production-informed, not demo magic
Written by someone who shipped code for 25+ years (ex AWS | Lululemon | Nike)
No hype, grounded in hands-on research and experimentation

Subscribe on Substack

Subscribe free →

Projects

A few things I've built alongside the writing. The newsletter shows how I think; these show what I ship.

Tilth

Active-Development

A harness for long-running agents that work autonomously, then let you replay the whole run.

A minimal harness for autonomous coding agents against any OpenAI-compatible endpoint. It carries the machinery a watched pair-programming agent can skip: an evaluator that judges whether a change is a proper solution, between-task budget caps, and offline-first observability you can replay end to end.

ResearchDev-ToolsObservability
PythonOpenRouterAgents

LSP4Agents

Active-Development

A language-server client built for coding agents, not IDEs.

An experiment in giving coding agents the structured tools developers rely on: go to definition, find references, real type errors. Existing LSP integrations assume an IDE on the other end, so this explores what a client designed for an agent's workflow looks like.

ResearchAgent-Tools
RustCLI

knobs.cc

A desktop inspector for every knob Claude Code gives you: where it lives, what it's set to, and which layer wins.

Claude Code's configuration spans settings files at several layers, environment variables, and permissions. knobs.cc is a local desktop app that surfaces every setting, its value, and which layer takes precedence.

Dev-ToolsObservabilityApp
TauriRust

Workshops

Live, hands-on sessions on Maven. You won't just hear about AI tools. You'll use them in real time, building artifacts you take back to work.

/ Always available

Free

Build the CLAUDE.md Your Project Needs

A 30-minute lightning lesson. Learn the framework for structuring the project context file that makes Claude Code actually understand your codebase.

Watch free on Maven

/ Upcoming

Packt × Deep Engineering: Effective Software Engineering with Claude Code, June 20, 2026
Hosted by Packt · Deep Engineering

Effective Software Engineering with Claude Code

June 20, 2026 · 10:30 AM EDT (4 hours)

From prompts to systems. For senior engineers, tech leads, and architects already using Claude Code who want repeatable results: CLAUDE.md context layers, reusable skills, guardrails, and team-level practices.

Register on Eventbrite

/ Previous

Maven course card for Context Engineering for Claude Code by Sam Keen

Context Engineering for Claude Code

Maven · 3-hour cohort workshop

Build the context layer that turns Claude Code from a suggestion engine into a development partner.

View on Maven
Sam Keen teaching a RAG workshop to a room of developers

Building RAG Applications

In-person workshop

Hands-on workshop teaching retrieval-augmented generation patterns with real-world datasets.

Sam Keen

Sam Keen

I've spent 25+ years shipping production software at companies like Nike, Lululemon, AWS, and a handful of startups. I led the GenAI Innovation Lab at AWS, where I helped teams separate signal from noise in AI adoption.

Now my work is AI-assisted development: building in the open, writing, and teaching. Building with AI, not building AI. My newsletter reaches developers every week, and my workshops on Maven give teams hands-on systems for working with tools like Claude Code.

I also co-founded Portland AI Engineers, a community of 1,300+ practitioners exploring practical AI together.

I occasionally take on select consulting engagements for engineering teams navigating AI adoption. Let's talk if that's of interest.