NDC AI Oslo · Room 3 · 10 June 2026

Take the whole drawer.

Everything from “What Actually Happens When You Give a Dev Team Claude Code for a Day” — the projects, the open-source tools, the papers, the write-ups. Build the one ridiculous tool, for an audience of one.

The slides → A live PR walkthrough → The 15 design themes →

Stian Håklev · builds developer tools at Tana, Oslo · @houshuang · LinkedIn · shaklev@gmail.com

i.

Built for everyone

Live sites anyone can use — each started as one person’s curiosity.

live site
Kulturbase

Kulturbase.no

1,500 Norwegian plays, 700 films, 1,300 concerts — searchable in one place.

→ live site
Koigen Hamar

koigenhamar.no

Every event in Hamar, in one calendar. Built in a day, for one town.

→ live site
MDG Programanalyse

MDG Programanalyse

7,916 proposals from 142 local party programs, classified by theme.

→ live site
Hirsch Argument Atlas

Hirsch Argument Atlas

10 books, 47 years, one argument — every claim mapped and cross-referenced.

→
ii.

Built for one — the stories

The two projects from the talk, in long form. These are the rich versions — start here.

The Alif story · Networked Thought

How an Iraqi Play About Cain and Abel Led Me to Build My Own Duolingo

Twenty years of wanting to read real texts, one too many spelling drills — and then one sentence to Claude, 43 research agents, and an Arabic learning system with calibrated sentences, my own data, and a cloned-voice podcast for an audience of one.

Read the story → The technical deep dive → Source on GitHub →
Alif — sentences, not flashcards
The Petrarca story · Networked Thought

Petrarca: An Intelligent Companion for the Self-Taught Reader

A mobile app that tracks what you know across books, podcasts, and voice recordings — and keeps it alive through contextual resurfacing. Think out loud on a forest walk; your wonderings come back researched, with sources.

Read the story →
Petrarca — voice recall
founded for one reader
Dragoman

Dragoman

A daily newspaper of the world in 12 languages — translated, clustered, laid out every morning.

→ the newsletter
Networked Thought

More where these came from

The newsletter where the drawer gets written up — tools, learning systems, experiments.

→
iii.

The dev tools — open source

The ones that followed me back to the day job. Take them.

$ review https://github.com/owner/repo/pull/123

Turns a 30-file PR into a guided walkthrough — reading order, architecture diagrams, the why — and collapses the trivial changes so attention lands where the risk is.

↳ see a real walkthrough of an Alif PR
$ /design-explorer "15 full themes, real data"

Generate fifteen complete design variants, pick by eye. It’s easier to choose than to describe — 15 → 1 winner in 31 minutes.

↳ scroll the 15 Renaissance themes
$ chat-search "database is locked sqlite WAL"

Semantic + keyword search over 8,000+ of your own Claude Code sessions — the tool that read its own logs and rebuilt itself. 18% → 0% error.
$ pip install limbic # the accidental library

Embeddings, search, novelty detection, clustering — extracted after every project kept reaching for the same code. I made a library and didn’t notice.
iv.

The papers & the framing

ESEC/FSE 2022
First Come First Served: The Impact of File Position on Code Review

Fregnan et al. — 64% lower odds of catching a defect in the last file shown vs. the first.

Radiology, 1990
Satisfaction of search in diagnostic radiology

Once you find one problem, you stop looking as hard. Old, human, and very much alive in code review.

CHI 2024
Data Storytelling in Data Visualisation

Shao et al. — narrative framing measurably improves comprehension of insights.

CHI 2025
The Impact of Generative AI on Critical Thinking

Lee et al., Microsoft — GenAI users self-report less critical-thinking effort; confidence in the AI drives the drop.

Essay
Simon Willison — “cognitive debt”

Plus his Showboat tool: make the agent write up how it verified its own work.

Essay
Addy Osmani — “comprehension debt”

The bill that comes due for code you didn’t read.

Write-up
How I Keep Accidentally Building Tools

The prompt → skill → tool pattern, in long form — on my newsletter, Networked Thought.

v.

The day job

Vibe-coding breaks at the team.

Tana puts AI agents inside the meeting, doing real work live — under the day-job rules: the AI never changes shared knowledge on its own. Every change is a proposal with a diff a human keeps, updates, or discards. Review, history, access control — the things developers take for granted, for everyone.

tana.inc → Aftenposten on Tana →
Tana — agents working in the meeting