Remote (US) · Full-time · Founding role

Founding
AI Engineer

Build the agent orchestration platform Lynton uses to deliver client engagements, deploy it on actual client work, and feed every lesson back into the platform. The feedback loop is what we're protecting.

About this role

Lynton has been building for the web since 1999. Sixteen of those years inside the HubSpot partnership. 2,000+ companies across 50+ industries. We left the partnership to build what comes next: AI-native infrastructure that companies actually own.

From inside Lynton we're incubating a new company. The model is service-as-software. AI agents not humans operate real business infrastructure for paying customers. The entity is taking shape right now and you'd be founding-team.

This role builds the system that makes the model work.

The role

You'll build the agent orchestration platform Lynton uses to deliver client engagements, deploy it on actual client work, and feed every lesson back into the platform. Same person doing all three. The feedback loop is what we're protecting.

We're past the point where "use AI tools to build systems faster" describes engineering work. Every developer does that now. This role builds the orchestrator, the playbooks, the autonomy gates, the evals, the bootstrap pipeline. Then you take what you've built into an active client engagement and use it. When it breaks, you fix it at the platform layer so the next engagement starts with a better version.

The platform won't get good without real client friction. That's why one person owns platform engineering and delivery in the same role.

You'll be the most senior IC on the agent layer through year one. The founder builds with agents daily and sets technical direction. You're the partner on architecture, not the executor of someone else's spec.

What you'll do

  • Design agent orchestration patterns: multi-agent graphs, tool routing, state and memory across turns, recovery from partial failure
  • Author and version the agent playbooks (discovery, content drafting, page builds, migration, QA, launch) that scale across engagements
  • Build evals, observability, and replay tooling so agent reliability is measured rather than guessed at
  • Design the human-in-the-loop interfaces for review queues, approval flows, and blast-radius gates
  • Engineer the bootstrap pipeline that turns "new client" into "live preview" in hours rather than days
  • Embed in active client engagements; the platform validates against real work, not staged demos
  • Make the design and product calls agents can't yet make well: typography, information architecture, brand interpretation, edge cases
  • Ship full-stack work directly when agents can't (custom integrations, bespoke logic, novel UX)
  • Capture every gap and failure from each engagement and turn it back into platform improvements
  • Version the template deliberately, with each release reflecting what the field surfaced

Who you are

  • 3+ years of production engineering, with 1+ year shipping agent systems specifically
  • You've shipped a multi-agent system to production. Not a demo or a chatbot side project: a real system with tool orchestration, state, error handling, and observability that runs unattended for paying users or business-critical workflows. You can describe what failed and how you noticed before the customer did.
  • Strong full-stack TypeScript/Node, ideally Python too. The platform is a full-stack product, and the agents output full-stack apps
  • Production reliability instincts: idempotency, observability, recovery, retries, blast-radius thinking. You ship things that run unattended in production, not things that pass tests on your laptop
  • Product taste. You decide what's shippable and what to cut. You don't wait for a spec
  • Comfortable embedding with customers. You'll be on calls when an agent's output needs human judgment
  • You build with AI coding agents every day (Cursor, Claude Code, Codex, Aider). Table stakes; we won't quiz it, we'll see it in your work

Bonus

  • +Production agent orchestration in any framework or rolled by hand. We care about what you shipped, not what library you reached for
  • +MCP server authorship or deep MCP integration work
  • +Comfortable across frontier APIs (Claude, Gemini, GPT) and open-weight models (DeepSeek, Qwen, Llama, GLM)
  • +Built dev tooling, CI systems, or developer platforms before. The agentic delivery system is one of those
  • +Worked in or near services delivery (agency, consulting, enterprise services). You've felt the cost of unreliable agent output when a client is paying

If you've only built demo agents, you'll be out of your depth. Same if you want to specialize narrowly. The role flexes between platform and client work constantly.

Our stack

Agent runtime

Custom Node.js/TypeScript orchestration, LangGraph and CrewAI where they fit, MCP for tool integration

LLMs

Anthropic Claude (primary), Gemini, GPT, open-weight (DeepSeek, Qwen, Llama, GLM)

Web

Astro, Next.js, React, TypeScript, Tailwind 4, MDX, headless CMS where it fits

Infrastructure

Self-hosted on Hetzner via Coolify, Docker, Caddy, PostgreSQL, WireGuard mesh, Cloudflare tunnels

Dev tooling

Cursor, Claude Code, Codex, Aider; Forgejo source control; CI gates for typecheck, lint, Lighthouse, axe-core, gitleaks, bundle budgets

Marketing & ops

PostHog, Listmonk, Twenty CRM, n8n, Plane, Matrix

The stack evolves. You'll have a strong voice in where it goes.

Why this role

  • Founding engineer for the new entity. We're spinning up a new company from inside Lynton. You'd be one of the first engineers on it, and the platform is yours to build.
  • You build the platform AND deploy it. Most AI engineering jobs are platform-only (you never see customers) or delivery-only (you never shape the platform). This one is both. That's how the platform gets good.
  • The platform is the moat. What you build behind the scenes is what makes the unit economics work. Anyone can rent the same models; very few can operate agents reliably enough to deliver on outcomes.
  • Real clients on day one. The platform validates against paying customers from week one, not pre-PMF prototypes.
  • A funded thesis with real demand behind it. Lynton has 27 years of clients and a credible book to draw from. You're building the supply side: agent-delivered services with software economics.
  • Founder builds with agents daily. Daniel sets technical direction. You're partnering on the vision, not interpreting someone else's.
Apply

No cover letter.
Just signal.

Four short questions designed to surface what resumes can't. We want to know how you think about agent systems, and what you do when one fails in production.

Applications are screened by an AI agent calibrated to the role criteria, then reviewed by a human. Specific answers beat polished ones, and AI-generated responses won't make it through.

If you make the screen, expect a 30-minute conversation with the founder, then a paid trial project on something real. No whiteboards or leetcode; we hire on signal, not theater.

Step 1 of 3

Tell us who you are

The basics. We'll only use this to follow up on your application.

City, state. Remote is fine, just helps us know your timezone

Step 2 of 3

Show us your work

Provide at least one: LinkedIn URL, resume PDF, or both. A portfolio link is optional but helps us see your design taste.

Required if no resume uploaded

Optional if you provided LinkedIn. Max 10MB.

Optional but recommended. Show us what you've built

Step 3 of 3

Tell us how you work

Be specific. Generic answers don't help either of us. A few sentences each is plenty. We're looking for substance, not length.

Specifics over polish. We can tell the difference.

A real example where you got the trade-off wrong, or where you got it right.

Observability, evals, recovery: pick a specific recent incident.

Link if possible. We care about ambition and outcomes, not titles.

Applications are screened by AI against the role criteria. Real humans review the recommendations.