Head of Product Engineering

Head of Product Engineering · Remote (US/EU) · Reports to the CEO
Immensity automates back-office work for operational businesses in construction, real estate and facilities: invoices, contracts, approvals, maintenance requests.
You own the path from a client’s messy workflow to a shipped, verified feature on Immensity’s AI agent platform. You decide what to build, write the spec, build much of it yourself with AI coding agents, and run the review and release process that keeps a multi-tenant system safe. You are the product owner and the engineering lead in one seat, so clients, engineers and founders all bring you the same question: is it built, is it live, and does it work?
You'll lead a small engineering team plus a fleet of AI coding agents. Also titled Product Engineering Lead.
What you will own
Product direction and design (~25%). Turn client calls, demos and feedback into specs, one-pagers and a sequenced roadmap. Own the information architecture and the product’s look, and make the build-vs-buy calls on models and integration vendors.
Build and fix (~20%). Ship features and root-caused fixes across a Python/FastAPI backend, a Next.js web app, an iOS shell and AWS infrastructure. Most of it through AI coding agents running in parallel, one worktree per ticket.
Quality, release and production (~20%). Test every release on staging and production in a real browser and simulator. Root-cause client-reported bugs, keep CI and deploys healthy, and get through App Store and Google OAuth security reviews.
Team and delivery (~20%). Write tickets engineers can build from, run sprint planning, review and merge PRs, and hire. Hold “Done” to verified in production. Own the AI development workflow: agent skills, shared memory, cross-model code review.
Client delivery (~15%). Stand up client workspaces and client-specific agents, prepare and run demos, and track each client’s usage. Be the technical voice on client calls and in weekly client updates.
Shares reflect how this work was actually split over the last 12 weeks, rounded.
What good looks like
By month 6, the delivery loop is one people trust:
Client demos run on the client’s real data and pass a readiness check before anyone joins the call.
“Done” means verified in production, and the sprint board matches reality without a manual audit.
Every PR gets a different-model review and merges without waiting on you.
By month 12, clients get value without an engineer in the room:
The redesigned product is the default for every client, and the old screens are retired.
A new client can sign up and connect their email and accounting tools with no engineer on the call.
Fixes stay fixed: each client-reported bug family ends with a root cause and a regression test in CI.
What you bring
We interview for each of these, and a clear no on any one of them is a no.
Full-stack range. Python and FastAPI, TypeScript and Next.js, PostgreSQL, AWS. Tenant isolation, secrets handling and audit trails are your defaults, not review comments.
Production scar tissue. You have owned something real users depended on, been burned when it broke, and changed how you build because of it.
Operates without permission. Handed an ambiguous problem, you narrow it, decide, ship, and say what you deliberately skipped.
Taste for slop. You use AI coding agents every day and catch the confident hallucination, the subtle bug and the plausible-but-lazy fix.
Fixes properly or files the debt. A shortcut under deadline leaves a ticket for the real fix, and you go back for it.
Comfortable in front of the client. You would join a call with a client’s CFO tomorrow and ask how their Tuesday runs before asking about their stack.
Direct, even when it costs you. You say what you think early, to the person who can act on it, with reasons.
Visibly faster than a year ago. You can name a tool or habit you dropped this year, why it lost, and what you are testing now.
Nice to have: shipped LLM agents with tool use, routing and evals; construction, real estate or accounts-payable domain knowledge; iOS App Store releases; SOC2 or Google CASA audits; hiring a small team.
How we work
Spec first. Every feature starts as a written spec, plan and task list, and ends as a PR to staging.
A different model reviews every PR. Authors never merge their own work. People review only security boundaries, hard-to-reverse migrations and money paths.
Client data stays in our cloud. Every model call goes through our own provider layer, secrets live in a managed vault, and every table enforces tenant isolation.
Done means verified in production. Merged is necessary, not sufficient.
Brief, factual writing. Updates go out by project, with no hype.
How to apply
Apply here. In your note, name one tool you dropped this year, what it lost, and what you’re testing now.

Work arrangement
Yes