Posts on lassence.net

From decks to one-off microsites

Sat, 17 Oct 2026 14:20:57 +0200

One benefit of AI that feels like a true liberation is that we can now move away from PowerPoint.

I have long been in search of an optimal solution that would allow me to focus on refining the content rather than polishing the slides layout. Manipulating blocks and arrows, changing font sizes and colors, can easily eat up more time than writing the content itself. That's why I like Markdown and Mermaid so much: you can just focus on your ideas, and let any renderer handle the formatting.

Building a 3D atlas of Paris as model benchmark

Sun, 11 Oct 2026 00:00:00 +0200

Public AI benchmarks are viewed with a lot of skepticism these days. That's why having personal benchmarks is useful: you test new models on the same problems and make up your mind.

So here it is, my own benchmark, like Simon Willison's famous "pelican on a bike", but more computationally expensive. I asked several models to create a detailed 3D interactive atlas of Paris, with buildings colored based on their construction period.

A personal data scientist for every employee

Fri, 09 Oct 2026 00:00:00 +0200

Notion shared a good article about how they gave their business teams AI agents to query their internal data:

How we built a personal data scientist for every Notion employee

It's a story that sounds familiar. In my experience, the data team is often a bottleneck for the many questions coming from business teams. Data analysts are a scarce resource, but they are the ones who know how the data is structured, and have the technical skills to extract it, then turn it into an analysis that (in principle) answers the operational teams' questions and need for insights.

Why I switched from Claude CLI to Claude Code for Desktop

Tue, 29 Sep 2026 00:00:00 +0200

While I used to think of myself as someone with deeply rooted habits in my work setup, after years of daily practice and tiny over-optimizations, it’s funny how I completely switched my work setup twice this year: in January and in September.

VS Code had been my daily work companion for years, first as a data scientist before the AI era, then when I discovered the power of Copilot completions, and later with more advanced AI agents. Until late 2025, I was still using agents like Claude or Gemini within the IDE, either in the dedicated Agents panel or in the integrated terminal. Up to that point, I wasn’t confident enough in AI-made code changes to trust them blindly: I had to check each change and work inside the files.

Will taste save us?

Sun, 14 Jun 2026 00:00:00 +0200

Anthropic released Claude Fable 5 this week, its new Mythos-class generation and a step above Opus. It pitched the model as “dangerous” so convincingly that, three days later, a US executive order forced it to pull access in a hurry.

A project I’d shelved for two years

I got to test it during those three days. Official benchmark scores track real-world use less and less, so I ran my own test instead. I gave it a project I’d been saving for a long time: analyze the meaning of a text and color each sentence by the sentiment it carries. The goal is a very visual rendering of literary or political writing, an emotional map of prose that sits somewhere between data science, literature and art. It’s the kind of project that needs a free weekend, and I so rarely have one that the idea had been sitting on a shelf for two years.

Which AI are we talking about?

Fri, 05 Jun 2026 00:00:00 +0200

For three years now, executive teams have had one order: do AI. But “AI” is a suitcase word. It packs several nested ideas: machine learning, deep learning, generative AI, and large language models (LLMs). People often mix them up, but when someone says “AI” today, they usually mean LLMs: ChatGPT, Gemini, Claude and their relatives.

Even if we stick to LLMs, there’s still an ambiguity. An organization can use them in two different ways:

We’re all becoming managers

Mon, 11 May 2026 00:00:00 +0200

With AI, most white-collar workers will change jobs without changing titles. They’ll become managers, of agents rather than people. The title stays the same, but where the day’s work actually happens moves a long way.

Knowledge work in three phases

Take any cognitive task you do on your own, roughly the work that happens behind a screen: data analysis, software development, legal drafting, market research. It breaks down into three phases:

AI isn’t the junior you were promised

Mon, 27 Apr 2026 00:00:00 +0200

The talented junior metaphor has run its course. Everyone reached for it to explain AI: a fast, technically capable assistant with no judgment, who needs a human to point it somewhere useful. A junior can hand you work that is technically flawless and strategically beside the point. The senior steers, reframes and makes the calls. The hierarchy held: AI executed, humans thought. That was true until 2025.

Say “technical tasks” and most people think of code, the go-to example. But code is only one slice. Recent models already match a qualified junior, at ten to a hundred times the speed, across a wide range of work: meeting notes, legal drafting, briefing memos, financial models in spreadsheets, first drafts of slide decks.