Sumit K (@sumitalk_s) header
Sumit K (@sumitalk_s) profile picture

Product Designer & Founder | Ex: Brandlive, Hubilo, Lendingkart | Building high-taste UI infrastructure and Figma systems for AI agents.

BLR
Joined August 2019
Photo by Sumit K (@sumitalk_s)Photo by Sumit K (@sumitalk_s)Photo by Sumit K (@sumitalk_s)Photo by Sumit K (@sumitalk_s)Photo by Sumit K (@sumitalk_s)Photo by Sumit K (@sumitalk_s)Photo by Sumit K (@sumitalk_s)Photo by Sumit K (@sumitalk_s)Photo by Sumit K (@sumitalk_s)Photo by Sumit K (@sumitalk_s)
Sumit K retweeted
The best part. You can just plug it into your backend. usetulip.co
Replying to @heydamir
This is the right instinct. Vibecoded dashboards don’t need more AI, they need taste. We asked #tulip to take on your Ember dashboard. We’re building the infra that makes a designer’s taste stick across everything an AI agent ships. Would love to compare notes.
1
31
The best part. You can just plug it into your backend. usetulip.co
Replying to @heydamir
This is the right instinct. Vibecoded dashboards don’t need more AI, they need taste. We asked #tulip to take on your Ember dashboard. We’re building the infra that makes a designer’s taste stick across everything an AI agent ships. Would love to compare notes.
1
31
token limit reset
1
66
Believe it or not this brew has 25g of protein!
1
1
159
Took a sloppy AI-generated dashboard and asked usetulip.co to give it some soul. Same layout. Same data. Now it actually looks like a product. Zero cleanup after.
1
184
Every startup is a restaurant. The kitchen (your backend) is spotless. Great chefs, great food. The dining room (your frontend) was decorated by five different people who never met. Tulip gives the dining room one look and hooks it straight up to your kitchen. Plug and play. usetulip.co
60
That’s the gap. Code generation got 10x faster. Context delivery didn’t move at all. The person with the taste is still copying it in by hand, one prompt at a time.
1
6
Better prompts won’t fix this. The fix is to treat taste as infrastructure: • Approved tokens and components in one source of truth • Streamed into the IDE over MCP • Checked at the PR, not caught by eye three weeks later
1
10
That’s what I’m building with Tulip, the design layer for AI-driven software. Human-approved tokens go straight into Cursor and Claude Code. The agent stops guessing. Private cohorts open now → usetulip.co
19
What actually happens on a client project: • Brand rules live in a Figma file nobody opens • Component decisions live in Slack threads • The real spec lives in one designer’s head The agent sees none of it.
1
11
So the designer becomes the context pipe: pasting hex codes into prompts, re-explaining spacing, catching invented components in review. On my last few sprints, I spent more time gathering and re-feeding context than designing.
1
5
I’ve been shipping production UI for clients as a #designer. Not handing off Figma files: merging PRs. The hard part was never the code. It was the context.
1
2
151
So it guesses: #F97316 instead of #F97518. A new <Card> when one already exists. rounded-xl where the system says 16px. Each one looks fine on its own. Together they’re drift.
1
12
Cursor and Claude Code changed who can ship. A designer with taste and an agent can take a screen from idea to merged in an afternoon. But every session the agent starts from zero. It doesn’t know your tokens, your components, or the 40 decisions you made last month.
1
42
Subconsciously being kind in my prompts because I want to be the last person they come for.
63
Introducing Tulip, the design layer for AI-built software. Tulip turns your brief into a governed design system that stays in sync across Figma and code, then pushes it straight into your agents via MCP. Taste infrastructure for a world where AI writes the UI. usetulip.co
1
1
6
341
Video editors on my TL. Let’s get some work done. Hit the DM
2
1
154
We decided to close down #Fozo, and honestly, it was one of the hardest decisions I've had to make. Building in this space taught me so much about execution, grit, and product than any textbook ever could. I’m deeply grateful to everyone who backed us, tried the product, and rooted for us along the way. Closing that chapter gave me the room to step back, recharge, and start building again with fresh clarity. Tomorrow, I’m sharing what’s next. Stay tuned.
2
8
404
Me jumping to conclusion
Turns out the future does have flying cars
1
62