hi bud buds,

for those that don’t know: cursor is my favorite product of all time. it’s an ai code editor that completely changed my life.

over the past 10 years, i spent thousands of hours trying to teach myself how to code, but i still couldn’t build anything more than simple apps. then cursor came along. suddenly, i could get working v1 versions of anything i could imagine, and just tweak them slightly to make them exactly what i wanted. it felt like magic.

if cursor has a million fans, then i am one of them

if cursor has only one fan, then that is me

if cursor has no fans, i am no longer on earth

if the world is against cursor, then i am against the world

- me

last week, cursor launched a big redesign called cursor 2.0, and i think it hints at how we might all use software in the future.

the big new feature: you can now have multiple ai agents work on the same problem at once. you can then compare their answers and pick the best one. this multi-agent approach apparently leads to much better results, especially for complex tasks.

when writing your prompt, you also choose multiple models to attempt the prompt

when cursor first launched, it felt like a normal code editor with an ai chat window on the side. a tool that ultimately you drove. now, it feels like cursor is rethinking that relationship. instead of us starting in the editor and using ai as a helper, we’ll start in ai chat and only use editors at the end for final touches.

said another way, i believe that in the future, we will:

  1. start with a chat prompt

  2. choose which agents to try the task

  3. review and compare the outputs

  4. make small edits yourself

i’ve mapped out this flow below:

i believe software platforms will need dedicated interfaces for chatting, reviewing, and editing

if you’re building products in this new ai era, a few key takeaways to consider:

  • let users run multiple models at once

  • build a review interface, dedicated to comparing outputs side-by-side

  • rethink your product’s role. if people only use your tool for final tweaks, things like search or navigation may not matter as much

have a great week, y’all. i love you.

- dj

last week’s biggest product releases

big news! huge!

personal faves

  • netflix announced interactive real-time voting, letting audiences of live shows vote as they watch. growing up during the rise of american idol, i absolutely love this format. although it do be feeling v black mirror

  • youtube shipped a tv app ux overhaul, revamping the interface to look more like netflix: big banners, thumbnails, and videos organized by seasons. i think youtube is wildly underrated in the streaming wars. first, gen z watches way more youtube than tv. second, as a social platform, youtube gets content 100× cheaper than netflix or hulu, who pay massive licensing fees

nice job guys

  • fitbit launched gemini-powered ai health coach, offering personalized guidance on fitness, sleep, and wellness

  • netflix launched redesigned profiles for kids, with simpler navigation and a new ‘my netflix’ hub for saved and favorite content

  • pinterest is testing ai-powered personalized boards, like creating outfits from saved pins and curating “boards made for you”

  • youtube released ai upscaling on low-res videos, automatically enhancing anything below 1080p. honestly, pretty sick

  • canva launched new ai features including form creation (a google forms rival) and email design tools for marketing and transactional templates

  • adobe released adobe firefly image 5, with support for layers and prompt-based editing, making it easier for artists to edit and customize images in their own style

  • adobe also released generate soundtrack and generate speech in firefly, offering ai-generated soundtracks. the soundtrack feature feels especially cool: it can scan uploaded videos and suggest vibe-based prompts (e.g. upload a chase scene and it might recommend a tense, edm-heavy track).

  • adobe also teased a chatgpt integration, allowing people to generate images, videos, and more using adobe’s models inside chatgpt

  • openai launched extra sora credits, letting users buy additional credits to generate more videos beyond their daily cap

  • threads released reply approvals, letting users control which replies appear on their posts

other llm news

  • canva launched their own foundational model trained on its own design elements that generates designs where each object and layer can be edited individually

  • ibm released granite 4.0 nano ai models, small enough to run locally in a browser

  • cognition released swe-1.5, a model optimized for software engineering with hundreds of billions of parameters

  • minimax released minimax m2, a model for agents and code, open-sourced on hugging face

idk tbh

  • grokipedia released grokipedia, an ai-written encyclopedia that elon is positioning as a “less biased” wikipedia. personally, some combo of current wikipedia + community notes feels like a better solution, as llms are still prone to bias

  • universal music group announced ai music creation tools for music composition using responsibly licensed training data. sadly, i would bet that models illegally trained on everyone’s music will outperform ethical models, leading to lackluster adoption

  • paypal announced a bilt integration, letting users pay rent or mortgages directly from venmo. i honestly just find venmo such a trash product that i have zero appetite to do anything more in there than i absolutely need to

  • ibm launched digital asset haven, a platform for financial institutions to launch blockchain services. i imagine ibm as a sad little clown with its pants down, chasing every trend, 12 months too late every time, tripping over itself, and ultimately never making any real progress in any direction

for my essentialism fans out there

that’s it for this week; thanks for reading. if you enjoyed this issue, please consider sharing it with a friend so i can get rich and build a home gym big enough that my friends could come over and we could workout together.

love, dj

Reply

or to participate