aeolian
2.9K posts
Claude Code in Figma isn't a feature. It's a hostile takeover of the entire dev handoff industry.
Every "design-to-code" tool for the past decade has tried to make developers' lives easier. Figma just said "what if we made developers optional?"
Here's what nobody's saying: the designer-developer handoff was never a technical problem. It was a political one. A negotiated border between two tribes with different languages, different incentives, and different ideas about what "done" means.
Claude Code in Figma doesn't build a better bridge. It annexes the territory.
The devs celebrating this as "finally, accurate specs!" are missing the plot. You're not getting better handoffs. You're getting automated out of the handoff entirely. That Figma file doesn't need you to interpret it anymore. It interprets itself.
The real winners? Designers who always secretly resented waiting three sprints for their vision to get "technically feasible'd" into something unrecognizable.
The real losers? Not developers. They'll just move up the stack. It's the entire ecosystem of handoff tools, design system consultants, and "bridging the gap" conference talks that just became solutions to a problem that no longer exists.
Figma didn't integrate Claude Code. They declared independence from your sprint planning.
Figma@figma
Tired: code or canvas Wired: code AND canvas Introducing Claude Code to Figma
English
Models are a bit more like people than many assume by default.
A post I see often is "Opus was stuck on this bug all day and then Codex solved it immediately, so I'm switching."
But wait.
In the old days when we wrote code by hand, this was an extremely common occurrence!
You'd be stuck on a bug for hours, finally call a colleague over, and they'd spot it in seconds. Because people have different strengths, and coming in with fresh eyes helps a lot.
The bug that Opus couldn't see is obvious to Codex with a fresh context window. But the reverse is likely true for a different task.
This, by the way, is why pair programming is so powerful.
It's way harder to stump two intelligences simultaneously. Their disparate strengths make them harder to block. They can catch their pair's mistakes, and sharpen each other by sharing hard-won lessons.
Hmm.
Thorsten Ball@thorstenball
@bretajohnson Replace "Opus 4.6" with "Senior Engineer #1" and "Codex 5.3" with "Senior Engineer #2" :)
English
@aeolianeth just switch to a vps lol. Vercel really isn’t needed for this, that’s the other reason why he’s funding this so that Riley won’t make a long post about how much they saved on their server bill and to switch off vercel
English
Gardening: New release of 🧿oracle is out, improving stability for using GPT-5.2 Pro from agents. github.com/steipete/oracl…
English
@thdxr @boristane yeah but then i gotta say "brah write the plan to markdown". Annoying. I guess it could be a skill?
English
so we shipped a more intense plan agent under an experimental flag
- i've barely ended up using it
- even when i tried it generated a big plan i felt annoyed reading/editing
- so now when the agent thinks to switch to planning im always like fuck no
there's some ux issues that can help but idk if i want to ship this
Darren Shepherd@ibuildthecloud
Sub agents and agent modes are crap. This is getting so dumb. I asked claude to implement a feature, so first it switches to plan mode, then it calls the explore agent, then it does some more random thinking and calls, then it calls the plan agent to create a plan. This is AI theater, none of this helped and so many calls were duplicated. This could have just been one linear thread. And then it by default want to "clear context." You know what the first thing it does after you clear the context!? It re-reads all the same files it just did. THIS IS SO STUPID AND WASTEFUL. This is what happens when engineers think they are smart. You don't need multiple agents, your just making it worse.
English
I spent a night this week pairing Claude Code with Linear’s new MCP tools.
Linear users create 2X more issues than legacy tools because they actually want to use it. That makes Linear a living, breathing context layer for AI coding agents.
The gap between the roadmap and the repo is closing.
Linear@linear
The Linear MCP server has been upgraded with new product management capabilities. See them in action in Claude Code:
English
@xrdevrob BRAIN-39. Grumpy model, huge context window and inference.
English
And here I am with my one MacBook screen and max. 1-2 codex instances running😩 What do people use to orchestrate all these different windows or agents?
Peter Steinberger 🦞@steipete
How the sauce is made @openclaw
English
@adamdotdev This “adult in the room” framing is pretty rude to the Claude Code team that built a product hitting $1B run-rate revenue faster than probably anything in history. Bun made like $2.50 total (stickers).
Engineering is relative to time & tradeoffs & they made fantastic tradeoffs
English
@nnnnicholas this guy himself will change his mind as soon as Google drops their next model 🤣
English
Opinions like this are so silly. Obviously Nano Banana Pro and Genie 3 are best in class. Obviously Gemini has best search index + youtube access. Everyone on AI X has a bag to sell you. Reminded of CT during boom.
Theo - t3.gg@theo
I really don’t see Google catching up any time soon. They’ve baked so much intelligence into their models and they are still so unpleasant to use. Beyond the model, the software matters more and more, and they are a decade behind there.
English
I am way happier with Linux than I expected to be. Almost all the issues I have are with hardware quality.
Mushy keyboard, flimsy screen with washed out colors, awful speakers, don't get me started on the trackpad...
English
Alex's first-of-its-kind accomplishment is a timely reminder that we humans are capable of unimaginable things.
AI progress is challenging our ideas of reality, creating a massive fear response that results in many feeling paralyzed.
Alex triumphed because he's trained his brain to silence fear. fMRI of his neural activity show a silent amygdala.
Achieving what you want in life can be done with a similar shift in neural architecture. What could you accomplish by overcoming default patterns such as silencing fear?
As a species, what if instead of destroying our enemies, we destroyed the concept of enemies. What could we then climb together?
That may seem impossible of humans, but as Alex shows, the risk is shorting our potential to change beyond our imaginations.
English
@ibuildthecloud @opencode most people don't seem to believe this though. Is there any real evidence either way besides bird app anecdotes?
English
There's really no reason to use Claude Code over @OpenCode. This is going to be interesting. The real value (as with all the this AI crap) is the model. You get the same model in both and opencode is just a better piece of software.
English