Original job description
Fully Remote · Occasional travel for team coworking and client visits · Reports to VP of Product · $110,000–$150,000 base + equity
About Your Role
Softrip is in the middle of a deliberate transformation - from a traditional SaaS travel technology company into an AI-first platform. Of our existing customer base, we have beta customers ready to build with us, solid discovery work done, and a clear direction. What we need now are dedicated people to build it - with external experience doing exactly this.
You’ll work across two products: one serving enterprise travel companies, one serving SMB operators - similar problems at different scales, giving you unusual breadth and real leverage. These aren’t two separate roadmaps running in parallel. You’ll identify where AI moves the needle in each and sequence accordingly. You’ll be hired alongside an AI Engineer as a deliberate two-person AI team - building together from day one on what to build and how to make it work.
The team is small by design. There is no separate design function, no QA layer, and no process infrastructure between you and shipping. If you’ve spent your career where shipping a feature takes eight approvals and three handoffs, this isn’t the right fit. If you’ve made decisions under uncertainty and shipped things yourself, keep reading.
The Job
Standard PM work - roadmap prioritization, customer research, stakeholder alignment - is part of this role. One sentence is enough to say so, because what makes this job different is everything else.
You own prompts and evals directly. Not conceptually - literally. You write system prompts, iterate on them, define evaluation criteria, build test datasets, and run them. When an AI feature produces bad output, you’re diagnosing why. You’ll live in Langfuse and LaunchDarkly as much as a roadmap tool. Your relationship with the AI Engineer is iterative and constant - there is no clean handoff between product and engineering here.
You prototype before you spec. When you have an AI feature idea, you test it before you bring it to engineering. That means stringing together APIs, prompting your way through a concept, or spinning up something in Lovable or Bolt to get to a real answer in an afternoon. The goal is to validate direction or kill a bad idea early - not to produce deliverables.
You design. There’s no dedicated designer on the team. You’ll produce wireframes and user flows using AI design tools sufficient for the team to build from. This isn’t about design craft - it’s about design judgment and the willingness to own it.
You review traces, not just metrics. When an AI feature misbehaves, you go into the traces - the actual sequence of inputs, tool calls, and model outputs - and diagnose what happened. Anecdotal feedback from customers tells you something broke. Traces tell you why and what to do about it.
You’re in front of beta customers. We have existing customers lined up to build with us. You’re in those conversations, gathering signal, translating it into product decisions, and feeding it back into the eval framework. The roadmap gets shaped by what you learn there, not just by internal conviction.
Critical Skills
• Prompt engineering at a professional level. Writing, structuring, and iterating on system prompts for real product features - not using AI as a tool, but shaping how AI behaves as a product. You understand failure modes and can improve outputs systematically.
• Eval ownership. You’ve defined what “good” looks like for an AI feature, built a test dataset, and run evaluations against it. You know the difference between a retrieval failure and a generation failure in a RAG system.
• Rapid prototyping. You’ve used Lovable, Bolt, v0, Replit, or direct LLM APIs to pressure-test an idea before engineering touches it. This is a reflex, not an occasional skill.
• RAG literacy. You understand how retrieval-augmented generation works - chunking, embeddings, retrieval, reranking - well enough to make product decisions about what context the model needs and how to get it there.
• Data fluency. You can pull and analyze data independently. You set dual success metrics: product outcomes and model evaluation metrics simultaneously, because one without the other isn’t enough.
• Small-team experience. You’ve shipped in a team of fewer than 5 people. You know what it means to make calls without full information, without a support structure, and without waiting for someone to tell you what to do next.
• Experience range: we’re hiring for capability and trajectory, not tenure. The right candidate might have 2 years of PM experience with significant AI features shipped in production, or 5+ years of traditional PM work with a serious, demonstrated pivot into AI building - show us what you’ve shipped. Side projects, portfolio work, and independent AI experimentation are legitimate signal and weighted accordingly.
Bonus Skills
• Experience with LLM observability tools - Langfuse, LangSmith, PromptLayer, or similar
• Exposure to agentic workflows: designing human-in-the-loop flows, understanding how agents take actions in a product context
• Former founder or consistent side project builder - you build things because you can’t help it
What Success Looks Like
First 30 days: Your first AI feature is live with beta customers. We have extensive documentation and direct customer access to accelerate your ramp - the expectation is that you’re making product decisions, not still getting oriented. You’ve run evals against the first feature and have a clear view of what to improve. You have a prioritized list of what to build next.
6 months: Multiple AI features are live across both products. The eval framework is running continuously, not as a one-time exercise. You and the AI Engineer have a working rhythm - you’re prototyping, they’re building, you’re both iterating. Beta customer feedback is actively shaping the roadmap, not just validating it.
12 months: Softrip’s AI capabilities are a material reason customers choose and stay. The roadmap reflects a compounding set of AI features - each one building on the infrastructure and learnings of the last - rather than a set of isolated experiments. You’ve helped define what it means to be an AI-first travel technology platform, and that definition is visible in the product.
