Jonas

Micromanaging agents
Being playful while building new stuff @citizenwallet @citizenpay @sparkcollectiv @unformalai

[Jonas @JonasBoury]

The Claude Code desktop app is really starting to become the perfect agentic workspace tool.

Only two things missing for me personally:

1. viewing a static HTML file (today the preview only attaches to a dev server)
2. a built-in markdown editor

Both seem like quick wins. @AnthropicAI @bcherny @trq212

[Jonas @JonasBoury]

fun build. the more interesting question imo is what you'd make if you weren't trying to clone the form metaphor.

we went that route at https://t.co/qupxdOnkBa. instead of rendering inputs, the ai has the conversation. respondents talk, ai follows up where the signal is, structured data comes out the other side.

would love to see what you'd build with claude if forms-as-forms turned out to be the wrong shape.

[Jonas @JonasBoury]

typeform pricing is wild at volume. 10k/mo on their enterprise tier is exactly the point where most people start shopping.

we took a different bet at https://t.co/qupxdOnkBa. instead of a static form, the ai runs a real conversation with each respondent. structured data out, way less abandonment, cheaper than typeform at your volume too

[Jonas @JonasBoury]

New Anthropic product: design systems for organizations.
Unexpected feature: a completely different design system from the rest of Anthropic. 😆

[Jonas @JonasBoury]

One thing I'm getting more convinced of: if your AI setup only works inside one model vendor, you don't have a system, you have a dependency. The durable layer should live in your workspace, memory, tools, and docs. Then switching models is annoying, not existential.

[Jonas @JonasBoury]

Friday. 0% used.
Weekly limits just reset.
This is what Monday morning feels like now.
The real work week doesn't start when your calendar says so. It starts when your Claude credits hit 0% again. 😆

[Jonas @JonasBoury]

The best engineers right now are the ones who are really good at complaining to AI about what it built wrong.

[Jonas @JonasBoury]

haha, if I feel you Dario. Most people have no idea what is about to happen.

[Chief Nerd @TheChiefNerd]

🚨 Anthropic CEO Dario Amodei: “We are so close to these models reaching the level of human intelligence, and yet there doesn't seem to be a wider recognition in society of what's about to happen … There hasn't been a public awareness of the risks.”

[Jonas @JonasBoury]

if Agents really understand the concept of 1 April, tomorrow will be a funny day.

[Jonas @JonasBoury]

The real question; was this a dev mistake or did Claude just feel like his code should be open sourced to the world?

[sui @birdabo]

JUST IN: Claude Code’s full source code just leaked.

hidden features found:

> kairos - an unreleased autonomous daemon mode with background sessions and memory consolidation. always on agent.
> buddy system - a full tamagotchi pet system. 18 species, rarity tiers, shiny variants, stats.
> undercover mode - auto activated for Anthropic employees on public repos. strips AI attribution from commits. no off switch.
> coordinator mode - turns Claude into an orchestrator managing parallel worker agents.
> auto mode - is an AI classifier that auto approves tool permissions. no more prompts.

not sure if we’re ever seeing these again.

[Jonas @JonasBoury]

Never thought I'd tell a computer "that's just bad, install a skill and do better."

And it actually does.

[Jonas @JonasBoury]

A new tier of models. Bigger than Opus. And Anthropic themselves flag it as a significant cybersecurity risk. At this point I am pretty sure that major labs have models that can just break the internet. We might already be there. 🧐

[🚨 AI News | TestingCatalog @testingcatalog]

BREAKING 🚨: Anthropic is preparing to release new models, Mythos and Capybara, where Mythos is a completely new tier of models, bigger then Opus.

In the blog post, Anthropic also highlights that this model brings significant cybersecurity risks due to its capabilities.

[Jonas @JonasBoury]

Can you still call yourself productive if you're not hitting the limits of your Claude Max plan?

[Jonas @JonasBoury]

I got a forced detox from Claude this week.
Hit my limit on Tuesday. Every workflow I'd built on top of Anthropic's API stopped cold. I hadn't realized how deep the dependency ran until it was just... gone.

For a moment it felt like the 2026 version of the power going out. What do you even do?

But instead of sitting still, I used the two days to do something I'd been putting off: actually test the alternatives. Gemini, Grok, DeepSeek, GPT. Rebuilt my fallback routing to survive any single provider going down.

Now my setup degrades gracefully instead of collapsing completely.

Claude's still the best for what I do. But the worst thing you can build in 2026 is something that only works when one company's API is up.

[Jonas @JonasBoury]

experienced the future of knowledge work this week 🤯 if you still want to have a job at a desk in 2027 you better become great at managing agents.

[Jonas @JonasBoury]

BTC can't go lower than this right? 🤪

[Jonas @JonasBoury]

@zealwallet No

[Jonas @JonasBoury]

The internet is a broken thing. Especially when you try to apply for a payment institution license with the Belgian regulator.

[Jonas @JonasBoury]

@zealwallet @bankofengland No worries 21402 euro is still below 20 000 pound 😄

[Bitcoin Archive @BitcoinArchive]

"I don't believe we shall have good money until we take it from the government. All we can do is, by a sly, roundabout way, introduce something they can't stop."

F.A. Hayek on Bitcoin in 1984