Forward-Deployed AI Engineer

Note

The job is a remote job and is open to candidates in USA. Truss is a fintech company building banking and payments products for construction businesses. The Forward-Deployed AI Engineer will own production internal AI systems, including the AI harness and fraud/customer-support back office, while improving engineering environments, agent skills, repository contracts, quality monitoring, and team processes.

Responsibilities

• Own the internal AI harness. It exists. You maintain it, harden it, and keep building it full time. Evals, tool wiring, sandboxes, whatever makes agents reliable here
• Own the AI back office. Fraud and customer support already run on an AI-built back office. You extend it, keep it correct, and sit with ops when a case flow is wrong. This is production, not a demo
• Make every engineer's environment actually work. Cursor, agent runtimes, virtualenvs, VMs, secrets, language toolchains. When someone clones a repo, the agent can boot, test, and ship. You notice when it cannot
• Own team skill sets. Shared skills for our stack and our way of working. You keep them current. Stale skills are a bug you fix, not a wiki someone might read
• Set the repos up for agents, not just humans. Devcontainers or equivalent, task runners, test entrypoints, lint, fixtures, docs the model can follow. The factory has a contract
• Build jobs that scrape the work. Daily signal on what shipped, review quality, test health, agent usage, where we are slow or sloppy. Improve the factory. Do not build a vanity dashboard
• Install process that sticks. How we open PRs, how we review with agents, how we promote a skill, how we kill a bad one. Boring, enforced, light
• Sit with the people who use it. Engineers and ops. You watch a session or a case fail and fix the harness, the skill, or the back office. You do not throw a Notion page over the wall

Skills

• 6+ years as an engineer, shipping production software
• A harness, an eval framework, an agent/skill system, or a meaningful open-source contribution in this area
• You have set up other people's environments, not only your own
• You treat harnesses, skills, and the back office like production code: versioned, reviewed, deprecated when they rot
• Comfortable in Python/TypeScript or similar: CI, scrapers, MCP/tool servers, sandboxes, ops UIs
• Small-team operator
• Must meet the location / work-auth rules in Sanity Check
• Remote (US or Canada)
• TN is OK if you are Canadian
• We are not sponsoring other visas
• You have been a working engineer for **at least six years**
• We do expect you have shipped real software for a long time, and that the AI/agent work is the reason you want this job
• A harness, an eval loop, an agent skill pack, or a real open-source contribution we can click
• You will not have a platform org
• You are the function for internal AI *and* eng acceleration
• You have already run a shared Cursor / Claude Code / Codex (or similar) setup across a team
• Published evals, SWE-bench-style loops, or sandbox/VM runners
• Contributions to Hermes, Claude Code skills, Aider, Continue, OpenHands, or similar
• You have shipped AI into an ops/fraud/support workflow, not only coding agents
• You have opinions on skills vs memory vs evals, and scars to match
• Optional: 3-8 minute Loom walking through the harness or OSS work (no secrets)

Benefits

• Equity 0.25% – 0.50%
• We can flex cash vs equity for the right person.
• Remote in the US or Canada
• Medical/dental/vision
• 401(k)
• Reasonable PTO
• Equipment
• Real ownership of internal AI (harness + back office) and how the eng team uses it
• A small, sharp team. Zero bureaucracy. High leverage.

Company Overview

• Truss is a financial technology company that moves money for construction businesses. It was founded in 2019, and is headquartered in San Francisco, California, USA, with a workforce of 11-50 employees. Its website is https://www.trusspayments.com.