AI and PRODUCT
RESEARCH for BOLD
VENTURES

Arc Labs builds AI-native products from research outward,
not the other way around.

The ventures, in production.

Each one started as a research question. It only became a product once the data said the question was worth answering.

2024Active

Snakke

Voice-native conversation practice

An AI speaking partner that trains real-time conversation fluency, not vocabulary drills. Built for the moment learners freeze mid-sentence.

2024Active

Dishy

Meal intelligence, not meal ideas

Reads your fridge, your macros, and your patience level, then plans what you actually cook this week, and adjusts when you don't.

2025In research

Notax

A tax copilot that reads receipts, not forms

Norwegian freelance tax is a maze of manual entry. Notax turns raw transactions into a filed return, and explains every deduction it took.

2026In research

Next signal

Under instrumentation

The next Arc Labs venture is being scoped in the lab right now. Follow the research notes below to watch it take shape.

The research log.

Most labs publish conclusions. We publish the log, including the notes that didn't survive contact with data.

Date	Note	Area	Status
2026-02-03	Latency budgets for voice-native tutoring loops Measuring where perceived responsiveness breaks down in real-time conversational AI, and what it costs to fix each millisecond.	Speech / Latency	Closed
2026-01-18	Receipt OCR failure modes across Norwegian vendors A field study of 1,200 receipts from Norwegian small vendors: where extraction breaks, and how confidence scoring should behave near the edges.	Notax / Data	Running
2025-11-29	Habit collapse in meal-planning apps after week three Most meal planners lose the user right after novelty fades. A model of why, and the intervention that held retention flat through week eight.	Dishy / Retention	Closed
2025-09-02	When to build vs. rent a foundation model A cost and control framework Arc Labs uses before greenlighting any venture that touches a foundation model in its critical path.	Venture Thesis	Logged

What we believe, stated plainly.

Most AI products are a feature wearing a company costume.

Intelligence isn't the moat. The workflow it replaces is. We only build where removing a task changes what someone can do next, not just how fast they do it.

Automation

Automation should remove decisions, not just labor.

Shaving minutes off a task is a feature. Removing the twelve small decisions someone dreads making is a venture. We build for the dread.

Education

Fluency is a nervous system skill, not a knowledge one.

You don't freeze mid-sentence because you forgot the grammar. Snakke exists because the fix for that isn't more content. It's more reps under real pressure.

Taxes

Compliance software should end at 'done,' not at 'here's your data.'

Most tax tools digitize the paperwork and hand the decision back to you. Notax exists to finish the job: filed, explained, and defensible.

Venture building

Diversification is a hedge against not knowing what you're doing. We'd rather know.

Arc Labs doesn't run a bet-on-everything portfolio. Every venture shares the same research infrastructure, the same standards, and the same founder's judgment.

Three ways into the lab.

Consulting

Short, high-signal engagements on AI product strategy, for teams deciding what to build, not how to prompt.

Partnerships

Co-building AI-native products with domain experts who own distribution but not the technical build.

Investment

Arc Labs occasionally takes early positions in ventures adjacent to its own research. Reach out with the thesis, not the deck.

Bring the thesis, not the deck.

Email hello@arclabs.no

Writing from the lab.

2026-01-22 · 6 min read

Why Arc Labs ships ventures instead of a portfolio of bets

Diversification is a hedge against not knowing what you're doing. We'd rather know.

2025-12-04 · 8 min read

The tax system was never designed for one-person companies

Notax started as a rant about VAT deadlines. Here's the research that turned it into a product.

2025-10-11 · 5 min read

AI tutors don't need to be smarter. They need to shut up sooner.

Notes from six months of Snakke usage data on when silence teaches better than correction.

Catherina, on the record.

I run Arc Labs, a Norway-based venture studio that builds AI-native products from research outward, not the other way around.

Before a line of product copy gets written, there's a dataset, a failure mode, and a reason to believe the idea survives contact with real users. Catherina.ai is where that research lives in public: the ventures it produced, the theses behind them, and the notes from the experiments that didn't make the cut.

Role: Founder, Arc Labs
Based: Oslo, Norway
Focus: AI-native product research
Active since: 2024

Let's build the next venture.

For ventures, research collaborations, or a founder-to-founder conversation, write directly.