Focus Areas

We explore the frontier of AI-built software — across voice, data, content, and custom applications. Here's where we're going deep. Not as a service offering, but as active exploration backed by real products.

AI-Native Software

What's Happening

AI agents can now build complete applications — frontend, backend, database, deployment — in days rather than months. But most of what gets built is fragile, insecure, and unmaintainable. The bottleneck has shifted from "can we build it?" to "should we build it, and is what we built actually good?"

What We're Building

Complete AI-built applications using multi-agent orchestration — each agent with clear context, responsibility, and constraints. Custom business tools that replace generic SaaS, shaped around how a company actually works rather than forcing adaptation to a generic workflow.

Explorations

Multi-agent build pipelines with specialized roles (architect, developer, reviewer, tester)
White-label business software tailored to specific company workflows
Architecture review and trust validation of AI-generated systems
Automated quality gates: security, scalability, UX

What we've learned: The engineers who struggle with AI agents are the ones who try to control how agents write code. The ones who succeed treat agents like team members — clear context, clear responsibility, focus on outcomes. It's remarkably similar to managing engineering teams.

Voice & Conversational AI

What's Happening

Voice is the most natural interface we have, and AI has finally made it real. Real-time voice agents can hold genuine conversations, handle complex workflows, and speak Nordic languages fluently. But building production voice AI — with proper telephony integration, GDPR compliance, and natural conversation flow — is still a frontier.

What We're Building

AI voice agents that make real phone calls — for eldercare, recovery support, customer outreach, and meeting transcription. We build end-to-end: from the AI model to the phone line.

Explorations

CallSia — Voice AI platform for Nordic businesses
Margit — Daily AI phone calls for elderly companionship
Refrain — Recovery support check-in calls
Proudfrog — Nordic meeting transcription, no bots needed

What we've learned: The most impactful AI isn't the kind you interact with on a screen. It's the kind that reaches people where they already are — through a medium they've used their whole lives.

Multi-Agent Orchestration

What's Happening

Companies are deploying AI agents across sales, support, operations, and engineering — but nobody's designed how these agents relate to each other or to human roles. Meanwhile, builders working with multiple AI coding agents face the same challenge at a different scale: how do you stay in control?

What We're Building

Tools, frameworks, and patterns for monitoring, controlling, and coordinating multiple AI agents. From development-time orchestration (managing coding agents across machines) to production-time orchestration (connecting business AI systems).

Explorations

Orchestrat.ing — Real-time monitoring and control of AI coding agents
Agent permission management and structured conversations
Multi-agent workflow design patterns
Integration architecture using MCP and A2A protocols

What we've learned: Managing AI agents is remarkably similar to managing engineering teams. Give clear context, define responsibilities, focus on outcomes — and resist the urge to micromanage. The best orchestrators think like CTOs, not coders.

Data Intelligence

What's Happening

Most interesting data is messy, scattered, and locked behind interfaces that weren't designed for analysis. AI makes it possible to collect, clean, model, and present data in ways that were previously too expensive for anyone except large enterprises.

What We're Building

Consumer-facing data tools that turn complex datasets into clear, actionable insights. We scrape, model, and visualize — making complex information accessible to normal people.

Explorations

Helanotan — Swedish car depreciation analysis from 200k+ Blocket listings
Multivariate regression modeling for consumer decisions
Automated data scraping and cleaning pipelines
Interactive visualization and comparison tools

What we've learned: The best data tools don't just show numbers — they answer the questions people actually have, in language they understand. If you need a data science degree to read the output, you've failed.

Content & Publishing

What's Happening

AI can generate content. Everyone knows that. What's less obvious is how to build reliable, multi-step pipelines that maintain quality at scale — with different AI models handling different tasks, automated translation, and human review checkpoints.

What We're Building

Complete content pipelines — from scanning and research to writing, illustration, translation, and publishing. Think newsroom-as-code.

Explorations

Hybrid LLM pipelines — Grok for research, Claude for writing
Automated translation to 6 Nordic languages
Daily AI news generation and publishing
Human review workflows with revision loops

What we've learned: The trick isn't getting AI to write — it's building the pipeline around it. Scanning, fact-checking, editing, translating, publishing. Each step needs its own agent with its own constraints. It's orchestration all the way down.

Something resonate?

We're always open to conversations with people exploring the same frontier. No pitch, no proposal — just a chat.

Get in touch