Thoughts on building with AI agents, serverless architecture, and the tools I'm creating along the way.
One plugin, two opt-in halves — transactional auth emails sent through your own AWS SES account, and signups synced to contacts with consent-safe defaults. Plus the hook that catches OAuth signups the obvious approach misses.
A typed Python SDK for AWS SES built on httpx and botocore signing — no boto3. Attachments, batch sends, SES templates, and suppression, with the same credential resolution as the JS SDK.
A better-auth plugin for in-app notifications. One plugin, one migration, one component — notifications live in your database, addressed to your users, with org-aware fan-out. No SaaS, no services to deploy.
React can't render HTML comments, so <!--[if mso]> is impossible. react-email-mso solves it in ~20 lines with custom elements and a string post-processor. Zero dependencies.
A CLI and library that validates email HTML against caniemail data for 30+ clients. React Email aware, CI-ready, with smart severity levels that don't drown you in noise.
How I build features with 4 parallel research agents, TDD-driven implementation, and adversarial code review — all orchestrated by a single skill.
A Rust CLI that catches when AI coding tools violate your team's conventions — banned patterns, deprecated imports, ratcheting legacy code, and Tailwind dark mode enforcement.
I built an email template editor with AI generation in 2 days using TipTap, React Email, and Claude. Here's how constrained AI generation made it actually work.