Every deal Groupon sells passes through one internal platform — Groupon IQ — where deals are created, checked and published. Seven engineers run it, and AI agents write much of the code. It's absorbing the company's other admin tools one by one; a deal lookup that used to take two minutes now takes thirty seconds. Nobody yet owns its reliability, or decides when agent-written code is safe to ship. You'd own both — and still be in the code.

What you'd actually get to build

Groupon IQ already runs on the company's modern engineering foundation — the platform the rest of the org is still migrating onto. That's your advantage: you build on the finished thing while making it load-bearing for everyone.

• One TypeScript codebase — ~95% of it, in a single repository, on the framework built by ex-Netflix engineers (Encore).
• Infra without tickets — declare a database, cache, or queue in a line of code and it's provisioned, wired, and observed for you: 60+ services stood up without a single infra request.
• Ship all day — every branch is a full staging environment; one pull request can change 23 services and two frontends and go out as a single deploy. Merges reach production the same day — minutes when it matters.
• A workflow engine (Temporal) runs the long processes — purchase flow, deal creation, outreach — reliably and auditably.
• AI is native — a shared model gateway and coding-assistant rules wired into every service. Engineers here average ~7 hours of real coding a day (versus ~3 elsewhere at Groupon); agents write real code, and review keeps it honest.

Who you'll work with

You'll report to Pavel Jakl (B2B Tribe Director), who writes code himself — a player-coach seat, not a calendar-and-1:1s one. You'll lead ~7 engineers across Prague and Bangalore in one codebase. Here, reliability is the product: as IQ absorbs more of the company's tools, more teams depend on it not breaking. Not for you if you want to stop coding, or want a pre-scoped backlog with someone else on the pager. Right for you if you like making a load-bearing platform faster and more reliable as it scales — and deciding how humans and agents build on it.

A week in the life

• Backend reliability, end to end — a long-running workflow stalls silently; you trace it through the platform's observability (no ticket, no waiting), fix it in one pull request, and write the post-mortem that stops it recurring.
• Production code, daily — real output, not a title.
• The AI code-review bar — recently an agent traced intermittent database timeouts to a call that never waited for its job to finish, moved every path onto the shared polling client, and added the tests that were missing: agent-written, human-reviewed, shipped. You own the standard that keeps work like that safe to merge.
• Absorbing the next tool — IQ is consolidating the company's admin tools; you make each one land without dragging the whole platform down.
• Cross-team architecture calls — you make them and defend them.
• Mentoring — seven engineers toward end-to-end ownership; feedback within a day, standards written down.

Signals you're a fit

• There's been a week where your commits and your team's output were both on the line — and both shipped.
• You've owned a platform where reliability was the product, and you moved the incident number.
• You've reviewed agent-written code and decided where the merge bar sits.
• You've traced a silent workflow failure end to end and written the post-mortem that stopped it recurring.
• You've automated away a manual task and spent the freed time building the next thing.

We'll dig into these live — bring the specifics. (You're still shipping, so show us something you built recently where agents wrote much of the code, and how you kept it correct.)

What success looks like

• Day 0–30: met everyone, mapped the incident history, shipped a real backend fix, started instrumenting a reliability baseline (there isn't a trustworthy one today).
• Day 31–60: the baseline is trustworthy; the AI code-review standard is live; one named slice of reliability debt closed.
• Day 61–90: incident rate measurably down against your baseline; an agent runs continuously, taking a manual task off the team; the next admin-tool migration lands without a reliability hit.

How we operate

• Extreme ownership — you own the fix and the post-mortem, not the ticket that reassigns it.
• Impact-obsessed — incident rate and coding hours are numbers you move, with a before and after.
• AI-first and disciplined — push agents to do more each month; hold the review bar that keeps them safe. Agency, not brilliant-jerk.

How we hire — 14 days to a decision

1. Recruiter screen (25 min) — career trajectory, role fit, comp alignment.
2. Foundry Challenge — a real-problem scenario from this role, done in your own time.
3. Hiring-manager interview (60 min) — walk through your Foundry Challenge, a deeper technical discussion, and how you operate.
4. Panel interview (60 min) — the rest of how you operate, plus hands-on technical depth.
5. Debrief & decision — fast; a yes comes with an offer.

Practical notes: Prague · standard employment contract · three days a week on-site · base salary Prague CZK 1,592,489–2,274,984 plus bonus · we'll ask your base location, work authorization, and the AI tools you use day to day.