Tyler.is ≡
@tylerwillis
Founder of http://Unsupervised.com (AI Data Analysis; insights found worth over $1B). I also invest in startups + build http://chatbetter.com (better LLM chat).
It's surprising to me how often simplifying the instructions you put in system instructions can improve LLM output. Giving detailed instructions is often fast and great for prototyping (and better than vague instructions), but getting it to be clear and simple is an art. It's a…
Solo founders are underestimated. We’re changing that. In May we launched the Solo Founders Program (<1% acceptance rate) for six founders to build together in SF. 60 days in: millions raised, 35k GitHub stars, rapid growth. Join us: Fall 2025 applications close soon.
Finally, we can own and control our messaging the way we own & control our money. Today at NewDayOne, @baseapp + XMTP unlock a new surface for builders: AI + Money + Secure group chats It’s time to build agents & mini-apps that work for people, not platforms. Let’s dive in👇
Hello from @baseapp Coinbase Wallet is now much more than a wallet. Your new home to create, earn, trade, discover apps, and chat with friends all in one place.
Here's a little inside information on how this all works, based a conversation I had with @ADoricko in California a few months ago. The key thing to understand is that Rainmaker's new technology has nothing to with cloud seeding. Augustus and his modestly-sized crew of nerds use…
I hope everyone is paying very close attention to what is being said here The CEO of Rainmaker, admits to not only cloud seeding in the area where the Texas Flooding is happening but it wasn’t 2 days before He says very clearly, they suspended operations THE DAY BEFORE the…
Yellow obelisk is an interesting design choice by the architect, but I am all for new builds.
BREAKING: Proposed S.F. tower would be 15 stories taller than Salesforce and forever change the city skyline. sfchronicle.com/sf/article/new…
"Nothing ever happens, because it happens so fast" is a really good summary. @balajis is masterful at concisely summarizing big ideas. Pace Layer Thinking made this make sense to me about a decade ago. Sharing in case it helps others. (link below) If you use Pace Layers to…
My answer is that we perceive rate of change but not absolute value. For example: there is far more usage of Uber today than 2014, but there was far more coverage of Uber then because it was new. In other words, we perceive the derivative of a sigmoidal happening process, but…
If you're looking for "Cursor for Slides" there are two good options right now: AIDE + Gemini 2.5 + MARP (markdown slides) PlusAI (Google Slides or Powerpoint) Both are really good options for very different user types.
Had dinner with the crew @solofounding a bit ago. Probably some of the most earnest + highest rate of learning people I’ve met in a long time. Something special.
The answer was OpenCode. The cost comparison is worth clicking through to see (I won't spoil it, the reveal is a fun moment of surprise).
Tried Gemini CLI, Claude Code and OpenCode on the same task on stream today. Guess which solved the problem in 1 try
This is the creator of @tailwindcss: - "didn't write a single character of code by hand" - "the code is exactly what I would have written by hand, no slop!" He heavily customized the rules file to accomplish this and is planning on releasing a refined version in Tailwind Plus.
Alright, really cooking now… ✅ Pixel-perfect compared to Figma design ✅ Supports light and dark mode ✅ Includes responsive design ✅ Didn't write a single character of code by hand ✅ Was definitely faster than doing it myself Figma MCP server + rules file biggest unlocks.
>80% of the conversations I have had over the past month with people building AI anything are about context. Really understanding how to build great context is a huge bottleneck right now.
Haha I'm not trying to coin a new word or something. I just think people's use of "prompt" tends to (incorrectly) trivialize a rather complex component. You prompt an LLM to tell you why the sky is blue. But apps build contexts (meticulously) for LLMs to solve their custom tasks.
Imagine suggesting to the CEO that your company give every developer a $2500/mo budget for using AI. If you think they’d react poorly, consider joining a new company.
In case you’re wondering why Cursor’s ARR is growing so quickly
the real lowest rate on the market is assumable mortgages from @laurenlself if ur buying a house talk to her
Some of you may have noticed that we're piloting a new mortgage benefit for our Robinhood Gold subscribers! A mortgage is among the largest and most important financial transactions of your life, and we're proud to be offering one of the lowest rates on the market, plus $500…
This is super well done. You should zero blind trust in media — AI has put the final nail in that coffin.
I’m biased positively here, but it also seems unequivocally true. It’s a good thing. You should use more AI. Even if it’s only to better understand the tools.
Imagine your CEO, CFO, and Head of Eng looking at a list of your company’s top AI coding agent users, and you’re at the very top in usage and spend. Is that a good thing or bad thing at your company?
AI does like 75% of the things I want it to do instantly and perfectly. The other 25% are a disaster. And I only really dare use it when I know the difference between the two. But the 75% is so valuable and fast it makes up for the rest of it.