AI Engineer; LLM Agents

Bytown's agents run on Anthropic's Claude through the Laravel AI SDK. Rule-based routing, backed by a Claude Haiku classifier, sends each chat turn to the right specialist agent, and we're rolling out an orchestrator and planner on Claude Sonnet for multi-step requests. Agents call tools that read performance and propose changes across Google Ads, Meta, Linked In, Microsoft and ChatGPT Ads, and stream their work to the app over SSE.

Customers can also run Bytown from Claude, ChatGPT and Cursor through our MCP connector, now in beta. You'd own that stack.

This is a production role, not a research role. Real customers run real ad spend through the agents you touch, and nothing goes live until the owner approves it. You'll iterate on prompts, build evals from real conversations, trade latency against cost on purpose, and make every proposed change clear enough to approve with confidence.

Most of the agent code is PHP. If your LLM work has been in Python or Type Script, that's fine — the hard part is the systems thinking, and Laravel is quick to learn.

• Own the agent framework end to end: routing, orchestration and planning, specialist agents and their tools.
• Build and harden tool calling into Google Ads, Meta, Linked In, Microsoft and ChatGPT Ads, with structured outputs the app can trust.
• Build evals before you build agents: regression suites, replayed conversations, LLM-as-judge where it earns its place.
• Ship prompt and model changes like code — versioned, reviewed, tested and easy to roll back.
• Own tracing and observability, plus per-agent cost, latency and quality. Use prompt caching and token budgets to trade them off deliberately.
• Grow our MCP connector so customers can run Bytown from Claude, ChatGPT and Cursor.
• Build guardrails so every change an agent proposes is accurate and easy for the owner to approve.

• Shipped LLM-powered features to real users in production — not just demos.
• Hands-on with tool calling, agents, streaming and evals.
• Strong backend engineering in any language — PHP, Python or Type Script.
• A measured view on cost and latency. You've cut a token bill or a p95 before.
• Opinions on when to prompt, when to retrieve, when to fine-tune and when to build a real tool.
• Comfort in a small remote team with no ticket factory.

• Built MCP servers or clients.
• Owned an LLM eval framework in production.
• Deep experience with Anthropic's Claude API, including prompt caching.
• Laravel or modern PHP.
• Side projects — open source, notebooks or public write-ups.

• Competitive salary + meaningful equity.
• A real API budget for experiments and evals.
• Remote — US & Canada.
• Direct access to customers — their conversations shape your evals.

Bytown is an equal-opportunity employer. We welcome applicants of every background, identity and ability, and we'll make reasonable accommodations at any step — just ask.