Engineering Leadership · AI
Software Development Manager, AWS · Amazon ECS
I've spent 20+ years leading engineering teams — from individual contributor through Director, CTO, and VP Engineering, across HealthTech, IoT, Oil & Gas, and cloud infrastructure. Today I'm a Software Development Manager at AWS, leading two teams building Amazon ECS — the software that powers the internet.
I started ashai.ca to write about engineering leadership and AI from the inside — field-tested on real teams and codebases every day, while completing my MSc in AI to sharpen the skills required for the next phase of our development journey. Everything here is learned in the trenches.
Every time you open Claude you make a choice you probably don't notice: how much of the work to hand over. There are three ways to work with AI.
Read → ai-leadershipDeciding how much to hand the AI is one move. There are four — and most engineers only practice the first two. Anthropic's 4Ds, in plain engineering terms: SSRR.
Read → ai-leadershipYour team is already using AI. They just don't agree on how. SSRR — Scope, Spec, Review, Release — gives engineering leaders a shared vocabulary for it.
Read →Most engineers open the .claude directory once, see CLAUDE.md, and move on. That's the whole mistake. Here's how to set it up properly — for a team, not just yourself.
Read → claude-codeCLAUDE.md loads at the start of every session. Get it wrong and Claude repeats the same mistakes forever. Here's how Boris Cherny — who built Claude Code — actually writes it.
Read → claude-codeCLAUDE.local.md never leaves your machine. Here's how to use it as a personal rule file built from the one source of feedback most engineers ignore — their own PR comments.
Read →The clearest AI foundations explainer I've found — neural networks through agents, no PhD required. I come back to this when I need to level-set with engineers still building their mental model of how these systems actually work.
The harness around the model matters more than the model itself. Cole Medin builds Anthropic's full playbook hands-on — CLAUDE.md hierarchy, self-improving hooks, path-scoped skills, subagents for exploration — and the self-improving stop hook alone changes how you think about keeping the AI layer from rotting over time.
Anthropic's official plugin directory — the right starting point before going anywhere else. First thing I did after reading it was check what equivalent exists inside Amazon's internal tooling.