Michelle

Mitul Shah (@typicalmitul, Design Eng at @vercel), Jamey Gannon (@jameygannon), and Cameron Collis (Design at @cursor_ai) showed a packed room of designers how they're working today. Mitul Shah (@typicalmitul), Design Engineer at @vercel 1:59 - Mitul kicks us off 2:24 - His design-to-production loop 4:49 - The redesigned Workflows table and trace viewer 7:12 - Why the trace viewer needs prod to keep iterating 9:32 - Yogurt, the Slack bot built to improve Workflows 10:19 - Q&A: annotations on images as instructions 12:40 - What dogfooding Yogurt taught him 15:00 - Q&A: trusting an agent you see through Slack 16:17 - Advice for designers becoming design engineers Cameron Collis, Designer at @cursor 17:29 - Cameron takes the stage 17:58 - The self-driving PR concept 20:13 - Lofi Figma exploration of the agent queue 26:08 - The custom prototype playground 27:57 - Walking the PR lifecycle inside the playground 30:27 - Q&A: why Figma is still his first stop 33:54 - Q&A: translation from playground to monorepo 34:52 - Q&A: standing up a prototype playground Jamey Gannon (@jameygannon), AI creative director 36:49 - Jamey closes us out 37:16 - Four principles for using genAI 39:29 - Aim before you shoot 41:50 - An image is worth a thousand words 44:52 - Nothing good arrives finished 47:17 - Use the right tool for the right job 54:31 - Q&A: reference libraries are the actual moat

@harmonic_ai 's engineering team is out in NYC next week talking Agents! Next Wednesday (5/20) @austinberke is joining @LangChain for a talk on Deep Agents and how our team is using them in our AI discovery tool, Scout. Luma: luma.com/nr72ijng?tk=0q… If you want to learn how we are building, or curious about working with the team at @harmonic_ai - I will be there :)

You might be overpaying 5.3x+ for Claude Opus 4.7! Our CEO @gabrielbianconi found out that on tool-heavy workloads, you're paying 5.3x more for Claude Opus 4.7 than GPT 5.4. The common metric is to compare cost per million tokens. But different providers use different tokenizers, and we found the same input produces wildly different token counts. Read our blog below to find out how you are being charged and how to actually know what you're paying for. 👇

Getting prizes ready for AI poker night with @TensorZero @BessemerVP 😈 on June 2nd DM if you’d like to attend !!

Spotted in the NYC subway. “Zero screen time.” An iPod Shuffle ad in 2026. When we built the iPod, the goal was the technology disappeared and you could have your music wherever you were. 1,000 songs in your pocket. Now we’re living through a moment where people are actively looking for ways to disconnect from the infinite feed, algos, and constant notifications. That doesn’t mean technology is bad. It means the best technology understands when to step back. Not every problem needs another screen, another menu, or another layer of complexity. Constraints create freedom (read: @DavidEpstein new book Inside the Box). And often removing features creates a better product than adding them. The future of technology shouldn’t just be more engagement. It should help us be more human.