The Opportunity 

Henosia has product-market fit. People are building real apps with us. They're automating work. They’re integrating systems and connecting data. We've proven that vibe coding works, not just as a toy, but for real businesses, in production. You can describe software in plain text and get production-ready code.

Here's the challenge: Creating AI agents that people can depend on to do business is super hard. The engineering challenge is to take something that’s inherently probabilistic, and engineer a system that makes it reliable across everything our users can throw at it. That’s where you come in.

You'll own the entire AI layer. The agent harness. The tool calling. The prompt optimizations. The eval frameworks. Everything that makes our agent smarter, faster, and more trustworthy.

You’ll take lead on this challenging work. No senior AI engineer to report to. It's you, the founders (who've been building software for 20 years), and a small team of full stack engineers in Denmark.

If you pull this off, you'll have built the brain of a product that changes how millions of people create software. That's the bet. 

What You'll Do

You'll own the AI layer. Specifically: 

• Build the agent architecture (harness, tool calling, prompt pipeline) 

• Keep up with AI research and adapt quickly when new models or techniques drop 

• Debug unexpected agent behaviors and address them at scale 

• Experiment with prompt strategies and model configurations to improve output quality 

• Build systems to evaluate agent performance and catch regressions before they ship 

• Work with the full stack team to integrate the agent’s capabilities and skills into the product (they write TypeScript/Next.js, you need to talk their language) 

• Ship constantly. This is a startup, not a research lab

Who You Are 

Your coding chops are impressive. You've been writing code for 5-10+ years, and you're good at it. You know TypeScript inside out.

You understand how LLMs work. Not just at an API level, but deeply. Prompt engineering, context engineering, function calling, model limitations, token economics.