I came across an amazingly intuitive explanation of durable objects on Twitter, so I put (ok, vibed) together a short visual explainer. Hope you find it useful.
I spent this morning with Martin figuring out a manifesto for Wasp - why we felt the need to create a new web framework and our vision for it. We've always wanted to have one in writing, and now it feels like the perfect time. Here's what we came up with:
Today I'm recording an episode of @_maintainable with @robbyrussell (you might know him as the Oh My Zsh creator) I'll tell the story of how my brother and I spent 5 years and $5M inventing a new language for web dev, admitted it was a mistake, and tore it out for TypeScript.
"conflate" is the word of the month here. Maybe even a quarter. I kinda miss "orthogonal".
How do you stay in the flow while coding with agents? Share your tips + tag folks who are really killing it in that department. I'd really like to figure this out.
Just realized Claude has its own system for subagents/async jobs/worktrees - how do you use it? And how does it work with orchestrators like Conductor/workmux/etc?
I'm looking for a few people to help us usability-test Wasp. The setup: hop on a ~30 min call, try building something small with Wasp while sharing your screen, and think out loud. I stay quiet and take notes - every "wait, what is this?" is what helps us make Wasp better.
We're looking for a designer to take Wasp to the next level. We just rebranded Wasp - blueprints/schematic style, monospace, three colors - and did the whole thing in-house. I'm genuinely proud of how far it's come. But I know a great designer could push it somewhere we can't reach on our own.
Achieving flow state with AI coding is still possible, it just requires more discipline. It's too damn easy to just start prompting before you've spent some time thinking first.
Martin just wrote a post that gives a name to what we're dealing with daily - fixing the AI slop. My favorite quote from Deslopifying:
People saying we don't care about frameworks anymore are wrong. People who cared about frameworks before LLMs still do. And there are more of them who got introduced to frameworks thanks to LLMs. You just don't hear them from all the folks who are now coding and don't know what a framework is.
Doing a task I don't really want (but have) to do is the best source of inspiration for me. I get a ton of ideas for tweets (like this one), Wasp features and new projects. I'm thinking I should strategically save such tasks for when I'm feeling out of steam.
How and where we talk about Wasp is just as important as the product itself. We noticed that over the last few months, as we were pushing the TS Spec release, we lost track of what we were doing and why on the content side.
"LLM writes better code than me" doesn't mean anything, and it's the wrong way to think about it. The best code is one that you and your team understand and can keep working on.
How do you decide when to use Claude chat vs Claude Code vs OpenClaw? I have all of them, and sometimes it's obvious, but often I'm not sure, and whichever I pick, I feel bad for the others who will miss that context. Any tips?
I see some people (even on our team) dogmatically resent letting AI write for them. Here's my experience:
One of my fav uses of AI is creating small throwaway "labs," as I call them - envs for trying something out. E.g., here I was designing a banner for Wasp's integration with @resend, and I was able to quickly go through multiple ideas and find one that I like.
One of my biggest regrets with Wasp is nothing on the product side. It's the fact that I haven't been more active here before. We always thought socials were a distraction (and they can be), but they are also a community of amazing, smart people you can tap into.
The best video about AI programming on the internet right now. Go.
In the end, the biggest growth "hack" is to build a product that delights the people who use it. That's all there is to it. I don't want to say "build the best product". That means nothing, and you'll never be sure you got there. Just delighting people is enough,
Two things that are driving people away from Wasp right now: - Deploying your app for free is not easy. We support one-command deploy to Railway and Fly, but you have to give your credit card - Our auth module is lacking - no single auth identity nor native passkey support
I feel that assembling your web dev stack today is like using Mongo. You get full freedom, no questions asked, but if you're not careful, the consequences will catch up with you.
Orchestrators are all fun and awesome, but for me, doing multiple things at once is still hard. I still have to load context every time I switch a project/task, and waiting for a prompt to resolve is kinda killing me. Any tips?
TDD: hey folks, I'm here, let's do it? Pre-AI: yeah yeah, ofc, ... (nobody does it) Today: OFC I'M WRITING TESTS TDD BDD WHAT DO YOU MEAN YOU READ CODE HOW DARE YOU x.com/MatijaSosic...
I miss the times when we argued about the best React form library. Btw, what are folks using today, react-hook or Formik? (giving it my best shot, let's see what happens)
My favorite @raycast feature since I became more active on X. Couldn't get anything done without it.
Anyone tried www.hunk.dev/? Or what do you rec for reviewing, if anything?
hunk — review-first terminal diff viewer
Hunk is a review-first terminal diff viewer for agent-authored changesets. Multi-file review stream, inline AI annotations, watch mode, and Git/Jujutsu integration.
hunk.dev
Not reading the AI code should be part of your toolbox, not a dogmatic worldview by which you define yourself.