Thomas Lemoine

Web Developer

[Wes Bos @wesbos]

the four horsemen of the apocalypse

[Carles Reina @Carles_Reina]

This is one of the coolest & fun use cases of our @ElevenLabs Agents platform. Charles Lorin built an AI Agent and called thousands of bakeries to know the price of a baguette. Well done!!

[Robert Bye @RobertJBye]

We need more fun app icons like these

[Enzo Manuel Mangano @reactiive_]

I've been recently experimenting with a new way to run e2e flows in React Native. Same maestro e2e test, same yaml, but 2-3x faster.

learn about Ennio 🧵

[Florian @flornkm]

Introducing the gradient border plugin

Familiar syntax, CSS-only, works everywhere. Easiest way for agents (and humans) to add border gradients via tailwind.

[Thomas Lemoine @thomaslemoine_]

WHICH AI? Guess the designer
https://t.co/XUo5FI7KoG

[David ImeI @DurvidImel]

2026

[Jon Cooper 🇺🇸 @joncoopertweets]

Okay, who did this? 😂

[Grzegorz WB @GrzegorzBk6810]

"Głowna różnica między ludźmi i zwierzętami jest taka, że zwierzęta nigdy nie pozwoliły by na przywództwo najgłupszego ze stada" - Winston Churchill.

[Grafikart @grafikart_fr]

Cette tendance à vouloir tous classer par "Most relevant" ça me fait vriller. Sur mon dashboard YouTube je veux voir les derniers commentaires, pas un commentaire vieux de 3 mois jugé "relevant" par un algo incompréhensible.

[nick @n___vc]

more people should use spinners like this, those are sick

[F.O.L.A @folaoftech]

Founder in 2026… doesn’t code 🫢😂

[Udayan Walvekar @udayan_w]

quick context on the problem 🙏
i have a loop setup with claude code and codex:

it builds something, takes a screenshot on it's own, compares it to the figma reference, keeps iterating until it's pixel perfect.

but it never works

the agent would screenshot its own output and somehow not see a completely broken button.

i'm sitting there staring at it! but under isay "the button is broken" it wouldn't see it.

i was like how are you so smart and dumb at the same time?

that's when it clicked for me.

the problem isn't that the model is dumb. the problem is that when you send a screenshot, the AI just... looks at it and describes the image in general sense.

so you spend 3-5 rounds saying "no, the padding is wrong." "no, look at the nav again." "no, the border radius doesn't match."

i started digging into how this is actually solved at companies that do it well.

turns out replit, lovable, microsoft, all of them solved this already. they all run structured analysis first. spatial coordinates, component hierarchies, design tokens.

then the model gets that as context.

but this intelligence is locked inside their platforms. if you're using claude code or codex or anything general purpose, you get none of it.

so i built clearshot 📸
open source skill for claude code/codex.

every time you or the agent takes a screenshot, it doesn't just "look" at it. it tells the model exactly what's there.

padding values in pixels. not "blue color" but the actual hex code. not "some spacing" but 8px gap between the label and the input. border radius, font weight, shadow values. everything the model was guessing at before, it now knows.

very deterministic but...
there's also a qualitative path. a taste layer. does the hierarchy feel clear. is the visual weight distributed right. is there enough breathing room. does this feel like a premium product or a hackathon project.

because sometimes the question isn't "what are the pixel values" but "does this feel right." clearshot handles both.

one thing i was careful about:
it shouldn't fire when it's not needed. if you send a screenshot of a chart, or a meme, or an architecture diagram, and you're not building frontend, the skill stays quiet. it only activates when the image is a UI and the conversation is about building or critiquing that UI.

and within the analysis itself, there are exit paths at every step. if a quick spatial scan is enough, it stops there. your agent isn't burning tokens running a full 5-step pipeline when you just asked "does this look right."

this is the first version. will keep iterating.
clearshot is open source and free.

star it if you think AI needs to get better at seeing your UI.

research this builds on:
- microsoft omniparser
- dcgen
- google screenai

[Thomas Lemoine @thomaslemoine_]

@kylian_khalifa RIVAL

[Thomas Lemoine @thomaslemoine_]

Pendant des décennies, l'Europe a misé sur les technologies américaines au lieu d'investir sur son territoire.
https://t.co/4xbLPk7oe0

[Jose @josesaezmerino]

Added cropping to Matte, and tried to do it in the most pleasing way.

[Andreas Storm @avstorm]

Gemini 3 Pro vs Gemini 3.1 Pro

[Claude @claudeai]

You can now push what you’re building in Claude Code directly into Figma.

With the latest updates to the Figma MCP server, build a working prototype in code, then send it to a Figma canvas to explore multiple versions.

[Tom Warren @tomwarren]

Anthropic just took a big swipe at OpenAI's decision to put ads in ChatGPT. Anthropic is airing ads mocking ChatGPT ads during the Super Bowl, and they're hilarious 😅 Anthropic is also committing to no ads in Claude https://t.co/LR1v4xz9ds

[Benjamin De Kraker @BenjaminDEKR]

Weekend AI coding project: Built a Perfect Popcorn Detection App🍿

Actively listens via microphone to popcorn pops in microwave, and tells you exactly when to stop.

Gemini + Claude. (Note: important to not actually place phone inside microwave, as this voids the warranty)