Why this role exists
HubSync builds the platform the top CPA firms in the country run on, and we are building AI into everything: the product (Halo, our agent platform) and the company itself. This role owns the second half: the AI-native SDLC, the entire lifecycle of how software gets planned, written, reviewed, tested, shipped, and operated here. We believe internal agents should be treated like product: an AI reviewer tuned to each repo's real failure patterns that earns the right to gate merges, a QA agent that tests every pull request against its ticket's acceptance criteria, agents that sharpen tickets and specs before an engineer touches them, that turn found bugs into fixes, that write the docs, that watch production and draft the RCA. The first of these are already running and the direction is set. What does not exist is the person on the ground making it real, because embedding with every team, building credibility engineer by engineer, contributing to their actual backlogs, and removing their actual blockers is a full-time job in itself. That is this role: the change happens because you are in the room driving it.
What you'll do
Build internal agents as products, across the whole lifecycle. Plan: agents that turn rough asks into well-specified tickets and keep specs honest. Write: the shared harness layer (skills, context, MCP servers, per-repo configuration) that makes every engineer's agent dramatically better than stock. Review: tuned AI code reviewers, codified review rules mined from our own PR history, and the measurement that lets them earn merge authority. Test: PR-triggered QA agents on ephemeral preview environments, exploratory bug hunters, generated deterministic Playwright suites. Ship and operate: CI/CD gates, deploy verification, incident triage and RCA drafting. GitHub Actions, GitHub Apps, Claude Code / Cursor and whatever tool is actually best this quarter.
Embed and convert, team by team. This is the core of the job. Sit with each squad and service, lead by example on their own backlog (pick up their tickets and ship them the AI-native way, visibly), showcase what is possible, and remove the friction that stops adoption: permissions, infra, environments, harness config, prompt and context engineering. Success is their throughput, not your demo, and it has to land as a better day-to-day for the developers, not a mandate they resent.
Own the measurement. Cycle time, review latency, escaped defects, cost per feature, and developer experience. The goal is a step-change in every team's throughput that the developers thems elves would fight to keep.
Make the standard real. Our doctrine: individual tool choice stays free, the shared layer does not. You build and maintain the harnesses and playbooks, wire the gates (no PR ships unreviewed or untested), sharpen the doctrine with what you learn in the field, and bring the taste to know when an agent is the wrong tool.
What makes you a fit
You are demonstrably, unusually productive with AI coding tools today, and you have already lived this transition somewhere: been there, done that, seen what works and what backfires. Not "I use Copilot": you run multiple agents in parallel, you have opinions about harnesses and context management, and you can show real PRs, tools, or automations from the last month that prove it.
Strong engineering fundamentals under the leverage: CI/CD, testing strategy, GitHub platform internals, infrastructure enough to unblock yourself (AWS), and the judgment to ship a working V1 in a day and harden it in week two.
You get more satisfaction from making a team 3x faster than from being the fastest person in the room, and you have the patience and credibility to win over skeptics with working software instead of arguments. Show, don't tell is how you already operate.
Comfortable being measured on adoption and dollars saved, not tickets closed.
What this role is not
Not manual QA management, and not a QA-only role: quality automation is the first battleground because that is where we are bleeding today, but the mandate is the full SDLC.
Not developer-relations or an evangelist seat. You ship, every week.
Not a research position, and not a decks-and-frameworks seat. The value lands inside the teams, on current-generation models and tools, every week.
Level
Staff or Senior depending on how much of the drive you can carry yourself: Staff runs the team-by-team motion across the org with minimal direction; Senior does the same work on fewer fronts at a time, with more of the sequencing done for them.
