Senior Applied AI Software Engineer

Kinter builds agents for mid-market finance teams. Prepaid, accruals, payroll, reconciliation, and more: the repetitive, judgment-heavy work that keeps finance teams stuck in spreadsheets every close.

We're a lean, product-centric team (~30 people) moving fast with an exciting product we launched a few months ago. Here you'll do the best work of your life and have an outsized impact: what you ship today is used by finance teams tomorrow.

How you'll make an impact here

• Evolve the AI harness that powers Kinter's agents: orchestration, tool-calling, context and memory management, and the scaffolding that sits between the models and production accounting workflows.
• Design and ship new agentic capabilities, from new agentic workflows to the structured interfaces that let agents act reliably in a financial system of record.
• Strengthen our human-in-the-loop patterns, confidence thresholds, and fallback behaviors that keep agents correct rather than merely fluent. In this domain, an agent that says "I'm not sure, here's what I'd check" beats one that confidently guesses.
• Expand the evaluation and observability tooling that lets us trust agent behavior at scale: curated datasets, evals, automated regression tests, tracing and logging so we can see why an agent did what it did.
• Work directly with customers and their finance teams to understand where agents can safely act autonomously versus where a human needs to stay in the loop, and translate that judgment directly into the harness.
• Help define Kinter’s agentic architecture as we expand from expense management workflows to a much broader set of agents; the patterns you build now become the template for what comes next.
• Work closely with product, engineering, leadership, and customers directly. Given the size of the team, decisions move fast and your work is visible.

What you'll bring to the team

• 2+ years of hands-on experience building and shipping autonomous agentic systems in production – not just prototypes, demos, or personal projects.
• 8+ years of experience building and shipping SaaS products. 
• Experience operating agents in a sensitive domain (finance, healthcare, legal, or similar), with a real instinct for why correctness is paramount there. You understand why human-in-the-loop beats a hallucinated best-guess GenAI response, the way it might for something lower-stakes like a marketing email.
• Direct, hands-on experience with agent frameworks (e.g. LangGraph, AgentCore, AutoGen, CrewAI, LlamaIndex) tied to a shipped product.
• Solid experience building web applications and services with Node.js and TypeScript, and comfort working across the stack as needed.
• Excellent problem-solving skills: comfortable diagnosing issues with incomplete information, breaking down ambiguous problems into actionable steps, and building resilient systems around inherently probabilistic components.
• A generalist and self-starter mindset. You're energized by learning a new domain (accounting) and by owning a feature from idea to production, rather than staying in one layer of the stack.
• Extreme collaboration. You'll work across small, cross-functional teams: knowledge-sharing, pairing with other engineers, product managers, and designers, removing bottlenecks and helping the broader product team hit their goals.
• Experience running production systems (deploying, testing, monitoring, debugging) and building with reliability and observability in mind from the start.
• Comfort with ambiguity and a startup pace. Our roadmap shifts with what customers need, and you'll help shape it, not just execute it.

How we work

Everyone at Kinter operates by five values. Here's what they look like in this role specifically:
• Be Fearless: You'll work in parts of the stack nobody has fully figured out yet, whether that's a new evaluation method today or a new agent workflow six months from now. That means moving into the uncertainty instead of waiting for someone else to derisk it first.
• Just Do It: We're a small team with a product that shipped a few months ago. A working agent capability you can iterate on beats a perfect architecture that takes a quarter.
• Sweat the Details: A misjudged confidence threshold or an agent that acts when it should have escalated can misstate a customer's books. Precision matters everywhere in this product, especially in the harness that decides when an agent acts on its own.
• Obsess Over Customers: You'll spend real time with customers' finance teams, understanding where they do and don't trust automation today, before you decide where an agent should act autonomously.
• Tell It Like It Is: If an approach won't hold up, whether that's an agent design assumption, a product decision, or a piece of scope you think is wrong, we want to hear that early and directly, not after it's live.

#LI-Remote