INTO AI is a Montreal-based company specialized in building and integrating AI systems inside complex enterprise environments.

We started by building our own AI SaaS platform for the hospitality industry. That is where our expertise comes from: running real AI products in production, with real users and real failure modes — not MVPs, not prototypes. We know what it takes to keep an AI system working after launch.

For over a year now, we have focused on consulting: helping companies deliver their AI projects and turn them into actual value. We don't just advise — we build and implement. Our systems run in our clients' own infrastructure, on their data, used by their people every day. When something breaks in production, we are the ones who fix it.

We are small and deliberately so. Every person here works directly on client mandates, and every person here talks to clients. Our team is already distributed across the world.

The surface you own

You own how an AI system behaves and how its quality gets proven, on client consulting mandates.

That means the production prompts, the acceptance criteria, the evaluation harness, and the demo where a client decides whether what we built is good enough. When the system is wrong 6% of the time, you are the one who decided that 6% was acceptable, measured it, and can defend it.

This is product management in a consulting context, not a roadmap PM role. No multi-year product, no backlog inherited from someone else, no market research. A client, a signed scope, a two-week cadence, and a system whose quality has to be demonstrated rather than asserted.

You work with an AI Engineer who builds the system, and a Delivery Lead who carries the mandate and the client relationship.

What you will actually do





Define what "good enough" means, and prove it. This is the core of the job. You build the ground truth set, design the labeling protocol, pick the metric and set the threshold. You do the reference labeling yourself, case by case — it is high-volume manual work and it is part of the job. You know when an LLM-as-judge is validated enough to trust, and when it is not.

Own the behavior of the system. You write and evolve the production prompts: the template, the input variables, the output schema, the model and its parameters. Not review them — own them. You define what the system does when it is unsure, slow or wrong.

Scope the work. You write the sprint scope every cycle, and requirements an engineer can build from without a follow-up meeting — including the data dependencies, the latency budget and the acceptance criteria.

Face the client. You run sprint demos for non-technical audiences, facilitate working sessions with their subject‑matter experts, and say no to out-of-scope requests with an argument that holds.

Build prototypes. Navigable prototypes covering real system states and error cases — shared with the client for validation before engineering builds anything, and with the engineer as the integration reference. You do not need designer craft, but a client has to take your output seriously.

What we are looking for





You have shipped at least one AI system into production and measured its quality after launch — whatever your title was at the time. Most candidates who fit come from an engineering background, product management, product ownership or data. In practice that usually means 3 to 6 years.

You have already had to prove that a probabilistic system was good enough — to a client, a regulator or an executive — and had the numbers to back it. That is the single strongest signal we look for.

You understand how LLM systems work — retrieval, structured outputs, agentic patterns, the quality/latency/cost tradeoff — and how they fail. You know the difference between a demo that works and a system that works.

You are technically autonomous. You will not write production code, but you read and write enough Python and SQL to automate a task, run a test and process evaluation data — with an AI coding harness doing the heavy lifting. This is why we favor an engineering degree or an equivalent foundation: if you need an engineer to test a prompt variation, the iteration loop that makes this role valuable disappears.

English is required. French is a plus — some of our client work happens in French. Internal communication, documentation and meetings are always in English.

That is the bar for a senior. We are also open to a solid intermediate with real potential: we have an internal training program and we invest in it seriously.

You'll thrive here if you...

Are comfortable being measured on the value you create. Not on velocity, not on tickets closed. On whether the system meets the bar you committed to — and whether it actuall

📌 AI Product Manager (Montreal)
🏢 Into
📍 Montreal