CAREERS · PM OF THE FUTURE · OPEN ROLE

The PM at Rebound is not what you think.

The legacy PM writes a PRD and waits three months.
The Rebound PM writes an eval and ships in a week.

Small teams. No cross-team dependencies. No process gates between a traveler's problem and a fix on their phone. At most companies, the same idea sits in a planning review, then a product meeting, then an engineering queue — and the traveler is still waiting.

PULL QUOTE · GEMINI HEAD OF PRODUCT

"Most PMs were never actually bottlenecked by execution. They were bottlenecked by taste and judgment. Team capacity functioned as a governor that prevented bad ideas from shipping. Remove that governor and you discover who was driving and who was just steering."

WHAT WE'RE LOOKING FOR

Trip Resilience Builder.

Not Product Manager. Not Senior PM. The title names the upgraded work — a role that doesn't exist at the incumbents.

We hire operators. They ship production code in Cursor or Claude Code on a Tuesday afternoon and run it against last night's real disruption logs before dinner. They write their own eval suites in Braintrust. They read a LangSmith trace without asking for help. They prototype the next Cascade Recovery feature themselves instead of filing it as a ticket.

Above all, they have taste — the judgment to know what's worth shipping when capacity is infinite, and the voice to tell a traveler "we think the 8am rebook is better than the 11am — here's why" instead of hiding behind "our AI decided." Beginner's mind on AI tooling. The stack changed last month; this builder already tried the new eval framework, the new observability tool, the new agent primitive — before the team asked.

THE OUTCOME THEY OWN · ONE NUMBER
% of disruptions resolved without the traveler touching their phone.
TARGET
87%
A WEEK IN THE LIFE

Prototype → evals → ship → review. Five days.

MON · APR 13
PROTOTYPE

Prototypes the rebooking agent v2 in Claude Code. Runs it against last Sunday's Frankfurt strike logs before lunch.

TUE · APR 14
EVALS

Writes 20 evals in Braintrust against last week's failure logs. Defines failure modes — hallucinated hotels, stale pricing, wrong currency.

WED · APR 15
SHIP

Ships the experiment to 10% of traffic. Opens a Linear issue with the eval run attached. No sprint, no review meeting.

THU · APR 16
REVIEW

Reviews eval deltas in LangSmith. Kills one branch. Doubles traffic on the one that beat the baseline on 17/20 scenarios.

FRI · APR 17
TALK TO USERS

Three Looms from three travelers who hit the failure mode. Watches them back. Files two new evals for Monday's build.

EXPERIMENTS / QUARTER
47
LEGACY PM AVG
3.6
LEARNING GAP
13×
PRDS WRITTEN
0
HABITS

Five legacy rituals this role refuses.

01
PRDs
Prototype in Claude Code instead. Evidence precedes documentation — if you can't show the working cascade recovery, a doc about it is just fiction.
02
Sprint ceremonies
The ritual was a governor on bad ideas. Taste is the governor now. No standups, no retros, no grooming sessions.
03
Stakeholder theater
One Loom beats four status meetings. The only meeting on a Resilience Builder's calendar is the one where someone looks at a live eval result together.
04
Handoff drift
No PRD → Figma → Jira → branch relay. The PM is in the codebase from day one, so the thing in production is the thing they built.
05
Yearly planning
Annual planning was a ritual for when changing course was expensive. Here the team pivots on Tuesday and the fix is in prod by Friday.
THE STACK

What a Resilience Builder lives in.

Explicitly not Jira, not Confluence, not roadmap decks in Google Slides. No feature has shipped from a Gantt chart at this company.

PROTOTYPING
Claude Code · Cursor
Prototype the agent themselves. Ship working demos in hours, not weeks.
UI / DEMOS
Bolt · v0
Spin up the morning-summary UI or a marketing page before anyone asks.
EVALS
Braintrust · Arize
Write evals alongside the feature. Catch hallucinations before travelers do.
OBSERVABILITY
LangSmith
Read traces of every agent decision. Know where the AI burns budget or fails.
ISSUES
Linear
Issues, not sprints. No point-planning, no velocity charts.
RESEARCH
Loom + a phone
Three user Looms beat a 20-page research deck. Watched raw, filed as evals.
THINKING
Claude · GPT · Gemini
Not a feature — a thinking partner. Fluency across foundation models is baseline.
BANNED
Jira · Confluence
Every hour spent grooming a backlog is an hour the traveler waits.
HOW THEY WORK WITH AGENTS

A day on the Resilience team.

They don't manage AI features. They delegate to agents, define failure modes, own the evals, and review the traces.

OVERNIGHT
The Pricing Agent shipped a 3% markup experiment against 14,000 active itineraries. Trust Agent scored 0.94 confidence — auto-commit.
09:00
Opens the eval suite. Spots that the agent has regressed on "traveler paying in a second currency" — 3 failures in the overnight cohort.
10:30
Writes 4 new evals in Braintrust capturing the failure mode — EUR wallet on GBP fare, USD wallet on JPY supplier, etc.
12:00
Ships a fix in Cursor that passes all 4 new evals plus the 47 existing pricing evals. No code review wait — the evals are the review.
14:00
Re-runs the experiment traffic. Watches the first twelve live cascades go green.
16:00
Reviews LangSmith traces for the afternoon cohort. Regression gone. Writes a 3-line Loom for the team. No meeting.
17:30
Talks to a traveler in São Paulo whose payment hit the bug this morning. Writes the apology message herself — doesn't hand it to support. There is no support.
Time from regression detected → fixed → verified: 7 hours
AT A LEGACY COMPANY · EST 6–8 WEEKS