I'm Jason Hartley, an independent senior engineer based in Austin, Texas. I build production AI agent systems and the full-stack infrastructure they run on.
Most "AI integration" projects fail one of two ways. They ship a chatbot demo that never makes it to production, or they bolt a model onto an existing app without thinking through state, retries, observability, or how a human stays in the loop. I work the other way around. I design the system first, the agents second, and the prompts last. The result is software that runs unattended, recovers from failure, and produces business outcomes you can audit.
What I build
Multi-agent AI systems.
I design and ship production agent architectures: orchestrated councils with specialized roles (strategy, analytics, content, finance, engineering), confirm-gate patterns for destructive actions, and shared-context layers that let agents coordinate without colliding. Recent work includes a five-agent operations council managing four parallel businesses, with each agent scoped to a defined domain and a human-approval boundary on every write.
Custom MCP servers.
I build Model Context Protocol servers that let Claude and other LLMs operate real APIs safely. One recent server connects an LLM to Meta Ads, Mailchimp, Shopify, and a Postgres content calendar, with a confirm-gate on every destructive action. The pattern is replicable for any business that wants their AI assistant to actually do things, not just talk about them.
Voice and conversational AI.
Production voice agents built on Twilio and Synthflow for senior living operators, integrated with their CRMs and lead-routing logic. Inbound qualification, callback scheduling, and human handoff handled end-to-end. This is the stack I reach for when a client needs AI to handle call volume their human team can't.
Full-stack TypeScript.
Next.js, Node, Drizzle, Neon Postgres, Shopify, WordPress. I've shipped a DTC ecommerce operation end-to-end (Kickstarter through fulfillment), a multi-tenant agency dashboard, and a board game collection management app. I run my own VPS infrastructure (Ubuntu, PM2, Caddy, Tailscale), so I know how the thing actually behaves in production, not just how it looks in dev.
How I work
A few principles I don't compromise on:
- Systems over scripts. If you ask me to do a thing twice, the second time I'm building the system that does it three times for free.
- Human in the loop on writes. AI agents require explicit confirmation for destructive actions. Always.
- TypeScript first. Strong types prevent the class of bugs that kill agentic systems at 3am.
- Boring infrastructure. Postgres, Node, Caddy, PM2. I optimize for things that still work in five years, not things that get a writeup this quarter.
- Ship something small first. A working v1 in two weeks beats a spec doc in eight.
Background
I've been shipping software professionally for over a decade and have spent the last several years going deep on AI-native systems. Today I run a portfolio of businesses end-to-end on infrastructure I built and maintain myself: a DTC ecommerce brand, a senior living SaaS platform, an AI automation practice for senior living operators, and a content and media operation. The day-to-day is operating at small-team scale using an agentic stack as the leverage point. That's the same stack I bring to client work.
Currently
Currently booking new engagements. Best fit:
- Companies that need an AI agent system designed and shipped end-to-end, not prototyped
- Founders who want one senior engineer covering infrastructure, backend, and frontend rather than coordinating three
- Teams adding voice or conversational AI to an existing product and need it production-ready, not demo-ready