Alexander Shatov

326 posts

Alexander Shatov banner
Alexander Shatov

Alexander Shatov

@alexbemore

product designer @instagram / @meta, gen ai team. building solo: https://t.co/fICeySDd93. UK global talent.

London, UK Katılım Mart 2020
783 Takip Edilen757 Takipçiler
Luke Knight
Luke Knight@lukeknight·
Coffee on me if someone in London can teach me about the thing they’re obsessed about. Anything - tennis, history, fine dining etc. Literally anything that you’re uniquely obsessed about. DM me. An hour on the weekend is perfect. I’m going @_sonith style
English
28
4
136
16.5K
Alexander Shatov
Alexander Shatov@alexbemore·
Amazing event with @glazeapp in London yesterday. Built a fully working AI chat app: Gemma local models. Native macOS UI. Fully offline. System prompts, folders, markdown… all in <2 hours. This is the future. Excited for @raycast’s Glaze launch 🔥
English
5
3
46
10.2K
Evil Rabbit
Evil Rabbit@evilrabbit_·
Next stop: London (for the first time).
English
20
1
145
7.3K
Benji Taylor
Benji Taylor@benjitaylor·
Rive + Claude Code is what I truly want
English
44
26
628
37.9K
Alexander Shatov
Alexander Shatov@alexbemore·
@alex_barashkov I clearly see this as one of new areas where designers can evolve right now. The era of design systems in figma has passed and designers now can not only support writing code for components in storybooks, but also build the tools like that. Amazing example @alex_barashkov!
English
0
0
1
454
Alex Barashkov
Alex Barashkov@alex_barashkov·
For more and more client projects, we build custom apps that transform uploaded images into a consistent, branded look. No more multi-step Photoshop workflows. We add the exact sliders, inputs, and options that make our work faster and easier.
English
13
9
276
22.7K
Alexander Shatov
Alexander Shatov@alexbemore·
Couldn’t believe that just 1 year ago I’d be able to build illustrations like this in code. Designers’ imagination has basically no limits right now with @cursor_ai.
English
1
0
3
277
Alexander Shatov
Alexander Shatov@alexbemore·
I do not chase perfect model setups to vibecode. I keep setup simple. Latest Gemini by default. Claude when Gemini hit limits. No obsessive cost optimization. No rigid rules. $60 cursor plan is perfect for almost all use cases.
English
1
0
0
217
joshpuckett
joshpuckett@joshpuckett·
As software gets easier to make, the products that stand out will be the ones crafted with uncommon care. If that's the kind of work you want to do, I'm sharing everything I know: interfacecraft.dev
English
348
265
5.6K
1.1M
Alexander Shatov
Alexander Shatov@alexbemore·
I never execute without a plan in @cursor_ai. Before execution, I always explicitly chose Plan mode, not Agent. That single decision removed most of the chaos I had before. Planning forces you to debug assumptions before they turn into code. Because models aren’t dumb, they’re eager. If you ask them to build, they’ll start immediately. And if they start from the wrong assumptions, you’ll spend hours fixing downstream effects. Plan mode interrupts that failure. The plan is visible. Editable. Nothing runs until I agree. I could remove steps, reorder things, mark things out of scope, and catch missing decisions before a single file existed.
Alexander Shatov tweet media
English
0
0
1
213
Alexander Shatov
Alexander Shatov@alexbemore·
I stopped trying to write “good prompts”. I also stopped trying to be clever with prompts. I stopped reading threads about “perfect prompting.” Strong models are already better at writing prompts than I am. Instead, I switched almost entirely to meta-prompts. I’d dump everything into one message: messy. unstructured. honest. Then I added one sentence at the top: “Write me a meta-prompt for ChatGPT that will…” The model will turn chaos into clarity. It will stop guessing implementation. It will structured my thinking. It will surface gaps early. Instead of me spending time figuring out how to prompt, the model will do it for you. At this point, ~90% of my prompts are meta-prompts.
Alexander Shatov tweet media
English
0
0
0
184
Alexander Shatov
Alexander Shatov@alexbemore·
I finally stopped being a @cursor_ai student and became a Cursor user. The first time around, I turned learning Cursor into its own project. Twitter threads. Perfect rules. Mega prompts. Folder structures. YouTube deep dives. Full workflows. I was preparing to build instead of building. This time, I limited myself to two links: 1. Eric’s internal Cursor guide. The same lightweight onboarding their own team uses. Minimal. Practical: x.com/ericzakariasso… 2. Lee Robinson’s kick-start walkthrough. How Cursor actually fits into day-to-day work: x.com/leerob/status/… That was it. About an hour and a half total. Enough to start. Here's the thing about consuming information: it feels like building. You're acquiring knowledge, organizing resources, getting ready. But ready for what? You're postponing the work while feeling productive. It's procrastination wearing a lab coat. We're good at learning. Learning is comfortable. Shipping is not.
Alexander Shatov tweet media
English
0
0
5
445
Alexander Shatov
Alexander Shatov@alexbemore·
5 years ago I published 28 icons on @unsplash. Simple 3d icons of popular apps. Nothing clever. Today they pull 11 million views per month. 500M+ views since I posted. Forbes uses them. TechCrunch uses them. AI can generate ten thousand icons before lunch. Yet editors at the world's most demanding publications keep reaching for my simple icons that I did in Blender. Why? I suspect it is because they work. They do the job without calling attention to themselves. They serve the reader. The lesson is old. Make something useful. Make it clear. Then get out of the way.
Alexander Shatov tweet media
English
0
0
3
199