AI Developer

About the role

This is the first engineering role we have opened at Sentry AI, which means the work is real and the ownership is real. You will not be handed a backlog of tickets. You will take agentic systems from a rough client brief to something running in production, and then own how it behaves once it is there.

We build AI for businesses that are counting on it: 15-plus systems shipped and every one still running, including a deployment serving 180,000 patients and a voice agent that screened 3,000 applicants in three months. The problems are genuinely state of the art, and they have real consequences when they are wrong.

We work AI first. Every engineer here builds with agents in the loop on current tooling. We are looking for someone who already works this way, not someone who needs convincing that it is the job now.

What you will build

• Design and ship agentic pipelines in Python: tool-using agents, MCP servers, retrieval over messy company data, and the evals that catch a regression before a client does
• Own FastAPI services that put those agents in front of live traffic, with tracing, evals and cost visibility wired through Langfuse from day one rather than bolted on later
• Take voice AI from prototype to production: latency, turn-taking, function calls into client systems, and the failure paths that matter when a real customer is on the line
• Write the MCP servers that let a model act on a business rather than just talk about it, with human approval gates where the stakes require them
• Work the full path when it is needed. Next.js, Supabase and Vercel for the control planes and interfaces that sit over your backend
• Run our in-house WATCH methodology on live projects, and sharpen it. It is new, and the engineers using it shape it

The stack

What we run today. You will not know all of it on day one, and we would rather you were deep in the backend half than shallow across everything.

• Backend. Python, FastAPI, Langfuse, Supabase and Postgres. Your centre of gravity. This is a backend role first.
• AI. MCP, RAG, agent harnesses, OpenClaw, evals, Claude and frontier model APIs. The agentic layer: how a model gets context, tools and limits.
• Frontend and platform. Next.js, Vercel, Supabase. Enough to ship the interface over your own service without waiting on anyone.
• How we work. AI-first engineering, agentic development loops, the WATCH methodology. Coding agents are standard issue here, not a personal preference.

How we work

Agentic by default. Coding agents, harnesses and context files are standard issue here. We expect you to arrive with your own way of working this way and to make ours better. The bar is not whether you use AI, it is how well you direct it and how quickly you notice when it is wrong.

The WATCH methodology. WATCH is our in-house development methodology for AI-first engineering: how a rough client brief becomes an agentic system that is safe to point at a live business. You will be trained on it and run it on real projects. It is new, so the engineers using it shape what it becomes.

What we look for

• Around three or more years building and shipping production software, with real backend depth in Python
• Systems you have actually put in front of users, and a straight account of what broke and what you changed
• You already build with AI agents daily: coding agents, harnesses, MCP servers, whatever is current. We will ask you to walk us through your setup
• Comfort with the awkward parts of AI engineering: retrieval quality, evals, latency, cost and non-determinism
• You keep up. Half the tooling that matters now did not exist eighteen months ago, and you already know what changed this quarter
• The right to work in New Zealand, and able to be in the Eden Terrace office two to three days a week

Nice to have

• Voice AI experience: Retell, ElevenLabs, or telephony in general
• TypeScript and Next.js beyond the basics
• Work in regulated environments: healthcare, finance, or anything carrying a real audit trail
• Open source, a technical blog, or side projects we can look at

None of these are requirements. If you match the list above and none of this one, still apply.

What you get back

• NZ$96,000 a year
• Hybrid by design: 2 to 3 days in the Eden Terrace office, the rest from home
• Current tooling, paid for. You will not be the engineer waiting on a licence
• Work that ships. Everything we have built is still in production
• A direct line to the founder, with no layers between you and a decision
• Real ownership: you design it, you ship it, and you own how it runs

How to apply

Apply in the dialog below. Five questions: your name, your email, how long you have been doing this, your CV, and a short video walking us through something you built. Or email james@sentrysolutions.ai quoting reference SEN-ENG-001. Lead with the work: a repository, a live product, an agent or pipeline you shipped, and a straight account of what broke and what you changed. That is worth more to us than a polished CV.

In the video, tell us how you use AI in your own engineering. Which harness, which agents, what you have automated and where you still do it by hand. That answer tells us more than a CV does.

Hiring is four steps and usually takes under two weeks. Your application, a thirty minute conversation with James, a short practical task that looks like the real job, then a decision either way with reasons.

See all open roles