benepass - Engineering Manager

• LEAD PEOPLE AND RAISE THE BAR • Manage ~4–6 engineers to start, with room to hire as we grow. • Run 1:1s that are about growth, not status. Set expectations against our IC leveling matrix. Give feedback early, including the hard kind. • Hire, onboard, and level the team. Design a hiring bar you would be proud to join. • Coach project leads. They drive kickoff, technical approach, weekly updates, and release — you make sure they have the support and air cover to do it well. • OWN TEAM EXECUTION AND OUTCOMES • Be accountable for results, not activity. Stay engaged until the problem is solved, not when the ticket closes. • Translate company priorities into a sequenced plan the rest of the company can plan around. If a timeline becomes unrealistic, raise it, cut scope, and make the tradeoff explicit — do not quietly slip it. • Protect operational load. Core Services carries production health, incidents, and correctness work that is not always a roadmap project. Plan for that instead of pretending every hour is feature work. • Incidents take priority over roadmap. Escalate risks early. • STAY TECHNICALLY CREDIBLE ON CRITICAL-PATH SYSTEMS • Lead design reviews. Champion simple, correct, observable patterns. Leave systems better than you found them. • Jump into code, prototypes, or production debugging when it unblocks the team or validates a hard tradeoff — then step back. • Hold the line on invariants that matter for money and eligibility: idempotency, auditability, reconciliation, and safe retries. Prefer a useful v1 over a perfect design debate. • Partner with Client Apps, Platform, Customer Engineering, Product, and Ops so Core Services is easy to build on — not a bottleneck wrapped in tribal knowledge. • WORK IN THE OPEN • Share context early. Document decisions. Surface risks before they become problems. Ask for feedback while the work is still easy to change. • Project teams are empowered to decide and expected to drive. Ownership includes gathering context, making tradeoffs, and unblocking yourself when you can. • BE AI-NATIVE — STARTING WITH YOU • This is not a “nice to have” and it is not a mandate you pass down while you keep working the old way. We want a manager who already heavily leverages AI in their own work, every week, across the job — coding, debugging, design, planning, writing, analysis, 1:1 prep, hiring, and communication. • Across Engineering, most new code is already written with AI in the loop. That is the baseline, not a future state. The managers who succeed here treat that as leverage: they get to higher-quality judgment faster, they automate the repetitive parts of the job, and they raise the team’s standard instead of lowering it. • What that looks like in practice: • You use it yourself, daily. You have a real workflow — not “we bought Copilot.” You can walk through how you actually design, debug, write, and review with AI. You have changed how you work, not bolted a chatbot onto old habits. • You hold the same bar. AI-assisted work is held to the same standard of correctness, security, and craft as everything else. On Core Services that is non-negotiable: ledger, decisioning, and money-movement mistakes are real. You review output. You do not outsource judgment. People own outcomes; AI does not. • You increase leverage, not ceremony. You look for work to automate: tests, runbooks, design-doc first drafts, incident timelines, onboarding, repetitive ops. You kill toil. You do not add an “AI process.” • You coach the team in the craft. Learning to work effectively with AI is now part of engineering. You set the expectation, share what works, and help people who are behind catch up — without turning it into theater or a purity test for one vendor. • You are curious and unsentimental. Tools change. You evaluate new ones the same way you evaluate anything else: does this make us more effective? You experiment in the open, keep what compounds, drop what doesn’t. • You do not hide behind it. AI is not an excuse for sloppy PRs, missing context, or skipped thinking. If the model is wrong, you still own the result.