In the last Artifacts, I argued that using LLMs to produce text may mean skipping the thinking that makes writing worth doing.

Yet, I’m definitely not a Luddite or sceptical of AI helping with daily tasks. It’s just that I don’t buy ‘the perfect prompt’ or other tricks with chatbots.

And so, while busy ranting against chatbots, I was also studying AI agents: artifacts that don’t just generate text on demand but can coordinate and use other tools, memory, and - more importantly - can perform actions.

And I started to wonder how I could use these artifacts to write Artifacts.

I moved from the tasks I do every day for the newsletter: pulling interesting articles or podcasts, storing them somewhere with labels or notes, finding moments to make sense of a lot of unshaped knowledge, but also trying to find that old article or books I read that may be relevant to a certain topic.

People working in product would say I started from the jobs to be done.

The more I researched, the more I realised that Claude Code was the perfect orchestra director of the different flows I was tinkering with. But let not be boring with the technicalities :)

What matters is that a few weeks in, I found myself with something I hadn’t expected at the outset: a full editorial infrastructure that autonomously gets relevant articles from multiple sources, organizes them and - when I decide on what to write about - supports me with the research and the refinement, tapping in everything I’ve read for the past 3 years.

All in all, I don’t think it’s an overstatement to say I got a small team of artifacts that help me write Artifacts.

Ah, a few important caveats:

• I use Claude Code, which is built for engineers, but works just fine for other tasks as well. A good alternative could be Claude Cowork.

• This whole infrastructure costs 19 euro per month.

• It’s hard to quantify how long it took, but it’s definitely easy to build something like this with AI, now.

Here’s a bit of the story of how it went.

First, the foundation

If there’s one thing that matters with AI agents, it’s context. Not copy-pasting the prompt of who you are, what you do, etc., but a more solid understanding and ‘memory’ of what you’re writing on, how you do things, what good looks like.

So when I started thinking about how to make Claude Code useful for Artifacts, the question was simple: where’s the best place to pull that context from? The answer was obvious: Artifacts itself.

I had Claude fetch and read every Artifacts I’d ever published - years of writing, including the early ones still in Italian, when the newsletter was heavier on policies and regulation and lighter on technologies.

And it was quite fun to ‘talk’ with the old newsletters, see how Artifacts has evolved and, implicitly, how I’ve evolved in writing them.

It was also the opportunity to rethink a bit how I describe this newsletter, settling on ‘stories & ideas on the design of technology’

Claude then used all this context to fill in the Claude.md, a file that loads automatically at the start of every session, where it’s clear who you are, what you do, where things are, what’s your style etc. Much better than the perfect prompt :)

Most developers use it to document code architecture. I use it to document Artifacts and how it is written.

CLAUDE.md is the most important piece in this stack. Everything else runs on top of it.

The second foundational piece is Notion. I’ve used it for years as a database where to automatically store all the articles, papers, podcasts, books, I read - basically, the knowledge that has been powering Artifacts for years.

Now, I’ve connected it to Claude (via MCP, for nerds out there), so that whenever I’m thinking of, say, age verification, it will return to me anything I’ve came across in the past 3 years, be it a podcast or a book highlight. Not too bad.

Claude.md and the Notion plug are where the full context lives: they don’t generate content but they make my agents aware of what Artifacts is and what I know already.

So I’m not always starting from zero, or chasing the perfect prompt :)

Then, the flow

Once the infrastructure knew Artifacts, I started using it.

First task is getting sources. As Artifacts is where I try to make sense of a lot of what I read, I try to read a lot. But as the newsletter subscription counter grows, it’s not the case for time - and sometimes I struggle to catch up.

So I built an automation in n8n that every day at 9 PM pulls articles from TechCrunch, The Verge, The Markup, and WIRED, plus whatever newsletters landed in my inbox, then filters, clusters, and analyses them with an AI tuned to ask one question: what matters here for Artifacts?

The output is some notes sent to a Notion page and in a Telegram message on my phone. I usually read it in the evening or first thing in the morning.

Sometimes nothing sticks, other times one item connects to something I’ve been turning over for weeks, and that’s when the idea of a piece begins.

Important point: I decide what to write about. For now, I don’t want AI suggesting topics. I enjoy connecting the dots myself.

Once I pick a topic, I open Claude Code and start looking into my Notion for pieces on the subject. It helps me re-read, draw an outline, ditch ideas, explore new angles - building the shape of what I want to write before I actually write it.

So I can draw on my notebook the mental schema I need before writing Artifacts.

Writing itself I keep for myself, and I think I’ve explained at length why :)

What I have built is/artifacts-polish, a skill that helps me clean language and keep it consistent with older editions, having ‘read’ the others.

In Claude, skills are fixed instructions on repeated tasks - say every morning you want a recap of your emails, you built a skill that knows what to do.

Once the draft is ready, I run this skill and it helps me go through what I’ve written with fresh eyes.

And when I’m happy, I send it out to your inbox :)

But it doesn’t end there. As some of you know, I always republish on LinkedIn - with some effort to avoid the annoying flattened language of that platform. So I built another skill,/artifacts-linkedin, that turns my piece into a post I’m actually happy with, based on all my previous ones.

Last but not least, Claude, the context and the agents keep learning from the process, so the infrastructure improves with me.

So, what?

I want to be careful not to oversell it. These artifacts don’t write Artifacts - but they help by surfacing topics I might have missed, pulling up pieces I read months or years ago, and keeping consistency across publications.

And I built this not to have someone else writing, but to do better some tasks and leverage more my very own work of the past years.

So it’s not just a writing assistant but more an AI-powered thinking and writing infrastructure, designed around what Artifacts is and has been, and with flows for how I want to be writing and keep doing it.

The critical decisions stay with me: picking the topic, building the argument, doing the writing. That’s intentional, and I don’t plan on changing.

But I must say that I’m having a lot of fun in having these artifacts that help me write Artifacts, and I’d encourage anyone to try it out for their projects!

Leave a comment

Save for Later

Why institutions are artifacts of cooperation.

What are the political effects of X’s algorithms.

Why we never asked for this Internet. And indeed we don’t really need these technologies probably.

What are the consequences of AI speaking without bearing the consequences of it. Btw, AI doesn’t like Naples.

Get bored, please!

The Bookshelf

As we discussed infrastructures, I recently wrapped up ‘Building and Dwelling’, a terrific book by Richard Sennett on how the cities we build shape how we live them. Not a tech book, but worth a read for those interested in Artifacts more broadly.

📚 All the books I’ve read and recommended in Artifacts are here.

Nerding

To complement work with Claude Code, but also to connect ideas and create your knowledge base, Obsidian is a good one. Needs an investment, but take a look!

☕?

If you want to know more about Artifacts, where it all started, or just want to connect...

Let's grab a virtual coffee ☕