All stories
MCP just got its biggest update yet: what it means for you
MCP is the standard every Claude connector runs on. A new version shipped today, the biggest since launch. Here is what actually changes in daily use.
7 design skills for Claude Code: how to improve any design
Claude Code has seven built-in design skills, from accessibility-review to ux-copy. What they do, when to use them, and the sentence that turns each one on.
Claude Design: from idea to working prototype in minutes
Claude Design turns a plain description into a real page with working code. What it is, who it helps, and how it differs from Figma or Canva.
Four things that break when you move to Opus 5
For anyone building on the API. Two changes throw a hard error, two others quietly cost you money. Plus the instructions you should be deleting.
The Opus 5 benchmarks, and the four rows it loses
Anthropic publishes fourteen benchmarks. Opus 5 wins ten. That is news, but the four it loses say more, and one number puts the whole table in perspective.
Effort is the dial that matters in Opus 5
Five settings, from low to max. The price per token has not changed, so this is where your bill gets decided. And the surprise sits at the bottom end.
I barely type to my AI anymore
I dictate almost everything, this article included. Allen Pike puts a name to my hunch: voice in, visuals out. Plus a few honest limits.
Claude Opus 5 is here: what changes for you today
Anthropic shipped Claude Opus 5 today. Same price as 4.8, close to Fable 5. What that means in practice, whether you just chat, build with it, or pay for it.
Anthropic put its best prompts in a library. Here's what I took from it.
Claude Code got a prompt library: copy-paste starters pulled from official workflows and internal teams. I didn't know it existed. Five patterns I picked up immediately.
Claude Sonnet 5's system card: what Anthropic tests, and admits
Every model ships with a system card: pages on safety, behavior, and welfare. Here's what stands out in Sonnet 5's: a model that criticizes its own constitution.
Expert prompting with Claude Sonnet 5
The close of the series, for the heaviest users. The approach of the maker of Claude Code, parallel sessions, and the choice between Sonnet 5 and Opus 4.8.
Advanced prompting: getting more out of Sonnet 5
You know the basics. Time for the controls most people leave untouched: effort levels, steering the length, and making the model check its own work.
Beginner's guide: giving Claude Sonnet 5 good instructions
Never really prompted before? Sonnet 5 follows you more literally than earlier versions, and that actually makes it easier to learn. The anatomy of a good instruction.
Prompting with Claude Sonnet 5: what you need to know
Since Sonnet 5 arrived, prompting works a little differently. Three things have genuinely changed. Here is what to know before you read the rest of the series.
A Chatbot Answers. An Agent Just Does It.
Agentic AI is the buzzword of 2026. Behind it sits a simple shift: the AI stops giving you answers and starts taking action. And that's exactly where the catch is.
Claude Cowork now runs without my laptop
Cowork leaves the desk. Scheduled tasks now run in the cloud, even when your computer is off. What was actually announced, how it changes my day, and where I stay careful.
Not Allowed at Work? Start at Home.
More and more people aren’t allowed to use Claude at work. At home you can, for free, on your phone. And that’s where you actually learn it.
Claude Fable 5: you're the bottleneck
At the AI Engineer World's Fair, Anthropic engineer Thariq Shihipar explained why you, not the model, are now the constraint. What Claude Fable 5 changes about how you work.
The plugin that figures out your Claude Code setup for you
Anthropic shipped an official plugin that scans your codebase and recommends which hooks, skills, MCP servers and subagents to turn on. It only reads. It changes nothing.
Dispatch: starting a Claude Code session from your phone
Dispatch can now start Claude Code sessions too, not just Cowork. You assign a task from your phone and pick the session up in the Code tab with Remote Control.
How Spotify lets Claude loose on 20 million lines of code
Spotify VP Engineering Niklas Gustavsson explains Honk, their agent fleet for code migrations, and why 73% of pull requests are now AI-authored.
Throw out the filing cabinet. Put AI at the center.
A Harvard case from the 90s explains why most companies see no gains from AI yet. Not by bolting it on the side, but by rebuilding your whole process around it.
Effort: the dial that decides how hard Claude thinks
Anthropic added a new dial to the Claude API: effort. You decide how deeply the model thinks, from a quick lookup to the heaviest coding job. Here is what it does and when to pick which level.
Karpathy calls Claude Tag the third redesign of LLM UI/UX
Andrej Karpathy says Claude as a Slack teammate is a new paradigm, after website and app. I read the thread and push back a little.
Everyone codes now. The finance team included.
At Claude Code the product manager writes code, designers build prototypes in the app, and the finance team runs its projections in a terminal. The roles are merging.
Claude Sonnet 5 is here: close to Opus, for less money
Anthropic shipped Claude Sonnet 5 today. It works more independently than Sonnet 4.6, lands close to Opus 4.8 on performance, and costs a good deal less. Here is what changes.
I do half my work from my phone
One of the makers of Claude Code starts half his agents from his phone, from the couch, coffee in hand. Not out of laziness, but because the work has changed. On working without sitting at your desk.
The bug was already fixed before I started
A colleague at Claude Code set up a routine that picks up bug reports on its own, posts a fix and pings the pull request. More and more often the work is done before anyone begins.
Verification is not your unit test. Can the agent run it itself?
Everyone thinks of tests and linters when they hear verification. The makers of Claude Code mean something else: can the agent run the thing itself and see that it works?
Tell the model less. It often knows a better way.
Giving an AI agent too much explanation looks careful, but it is a form of micromanaging. The makers of Claude Code call themselves context minimalists. Why saying less works better.
I dropped Plan mode. Now I do everything in Auto mode.
Boris, one of the makers of Claude Code, no longer uses Plan mode. Only Auto mode. I work the same way now. On a habit that nothing needs anymore, and the one reason I sometimes keep it.
Claude Code gets Artifacts: share a working page from your session
Anthropic brings Artifacts to Claude Code. Build an interactive page from your session and share it with your team through a private link. Now in beta on Team and Enterprise.
Getting started with Claude Cowork
A practical guide to get going with Claude Cowork in one evening. From your first task to projects and a custom skill that takes the dull work off your plate.
Claude designs with me now, and it talks to my code
Claude Design went into beta today on every paid plan. I tried it right away and it works great: more designing, and a smooth handoff to Claude Code.
Why I Do It This Way
I could have made it easier. A subscription, a button, done. Yet I make my images locally and in a single style. About that choice, and the lessons it taught me.
How to Make Them Yourself
A practical guide to making images locally with Bonsai. Install, one command, and the secret to holding a single style. Plus a gallery of what is possible.
Builder Day in San Francisco: what seven teams built in a single day
Anthropic ran a Builder Day in San Francisco. Robert Scoble posted the seven best teams. I watched their presentations and pulled out the insights and lessons.
Fable 5 pulled: the government withdrew the strongest model
Three days after launch, the US government suspended Claude Fable 5. What happened, why Anthropic disagrees, and what it means when you build your work on a model like this.
Images Without the Cloud
The images on this site are made on my own laptop, offline. No data center, no bill per picture. Here is why that is suddenly possible, and why it matters.
How Hard the Machine Thinks
There is now a dial for how deeply the machine thinks. Since Opus 4.8 it sits on high by default. How I decide when to open it up and when to leave it alone.
Test-driving Claude Fable 5: an afternoon of real work, not a demo
I handed Claude Fable 5 my full content production for one afternoon: two campaigns across eight systems. How the new model feels, what it costs, and what stayed human.
Claude Fable 5: the strongest model you can use today
Anthropic shipped Claude Fable 5 today, the first publicly available Mythos model. Here is what it is, why it feels different from a normal release, and what it changes about how I work.
The prompting playbook: let Claude write your prompts
The best prompt is not one you write yourself. You let the model draft it, you test, and you refine. What that changed about how I build with Claude.
What's New in Claude Opus 4.8
Anthropic shipped Claude Opus 4.8. I use these models every day in sessions and build with them myself. Here is what stands out, and which options I actually turn on.
Claude Cowork: a month of double hours
Through July 5, Anthropic doubles the 5-hour usage limit in Claude Cowork. No button, no request. Here is what it means, and how I use Cowork every day.
One word spun up a fleet of AI agents. Now you have to say it on purpose.
A single word quietly launched a swarm of subagents in Claude Code. Anthropic just made that trigger explicit. Small patch, large principle.
Eight Times the Code, and What It Doesn't Say
This week Anthropic put a number on something I had felt: eight times the output. The number is probably right. The question is what it measures.
What Claude Can Do Now That It Couldn't Last Year
Not a spec sheet, but how I actually use the newest capabilities: connectors, background work, skills, and a homegrown image pipeline. And where I still keep my own hand on it.
The Assistant That Executes
As long as AI lives in a chat window, the advice stays at a distance. Now that remote connections are on by default, the assistant moves into your tools. That is not a feature, it is a shift.
How I Left WordPress Behind
No more plugins, no page builder, no login screen between me and publishing. I now build and run my sites entirely with Claude. Here is what changed.
AI Psychosis and the Last Mile
A tech CEO called his peers prone to AI psychosis. He has a point, and it comes down to the distance between the boardroom and the actual work.
My brain doesn't fry
Research calls AI brain fry a new workplace crisis. With ADHD and a conceptual brain, I feel none of it. Why my cognition is only now falling into place.
Digital Sovereignty Isn't About Technology
After the Sovereignty & Government event, I understand better why this debate keeps stalling. It's not about bits and bytes.
Open Innovation Doesn't Happen on Its Own
Esther Gons told me how open innovation keeps stalling inside large organisations. I recognised every word of it.
What a Label Tells You
Wine and photography overlap more than you'd think. Both demand that you look closely. A label never lies outright, it edits.
Rules as Code: why legislation should be readable by computers
In March I attended the Rules as Code conference in The Hague. What I learned there cuts to the heart of how government works.
Five Grower Champagnes Worth Discovering
No big houses. Five small producers I open often, and the ones I hand to anyone who asks where to start.
Champagne Is Not a Party Drink
I drink Champagne on a Tuesday night, with pasta, for no reason at all. Here is why that turned out to be the best wine decision I ever made.
The Synthetic Sommelier
What does it mean when AI recommends wines better than a human expert can? An exploration of taste, algorithms, and what we lose along the way.
Why I Photograph
Photography and wine don't look alike. But they ask for the same thing: looking slowly.
Why AI Strategy Is Always About People
Organisations pour millions into AI projects that fail. Not because the technology falls short, but because they forget the people.
The Camera I Never Put Down
On photography as a way of seeing, the discipline of throwing away, and why a good photo is rarely the moment you thought you were shooting.
No stories in this category.