I write a newsletter about startups and investing—for ai builders of all levels.
I record mini-tutorials, review tools I’m testing, share my insights and give you a peek behind the digital curtain from an exited founder turned investor.
how do we make info easier for llms to understand?
you may (should?) already know about xml tags in your prompts to help set context, e.g.
Write a summary of climate change. Format your response using these tags:
<causes>List the main causes here</causes>
<effects>Describe the major effects here</effects>
<solutions>Outline potential solutions here</solutions>
now people are making llms.txt files - a simple markdown of your website content, see an example from the AI engineer website.
why? because it’s easier for llms to understand. dev tool documentation use it too.
i read this yesterday: “The value of llms.txt: Hype or real?”
this is the tip of the iceberg.
it feels as important as setting up your title/description/etc for search engines. and i think there’s so much more to explore.
i think we’re going to see a wave of tools, systems and startups that simplify for llms.
we work for the llms now, we should be thinking how do we serve them the best info in the best format, and get out of their way.
who’s tinkering in this area? send me a message.
come and talk about what you’re building, excited about or tinkering with!
p.s. we’re going to do a hackathon soon, lmk if you wanna join
🔎 News worth knowing
OpenAI is rolling out GPT-4.1 in ChatGPT starting today. Also, GPT-4.1-mini is set to replace GPT-4o-mini. GPT-4.1 family was earlier released for developers only in the OpenAI API, but the model is a really good workhorse with 1M token context window (OpenAI’s first such model). Here’s a handy GPT-4.1 prompting guide from OpenAI.
Keep an eye on these:
Rumours are swirling about Anthropic's new Sonnet and Opus models having more advanced reasoning.
Hearing whispers about MCP connectors in ChatGPT potentially coming soon.
The Information suggests Google might launch an always-on dev agent at I/O.
Gemini + GitHub is live. All major AI apps let you connect to github now. Canvas in Gemini also got updated, allowing users to build on others' canvases and enabling apps built in Canvas to use your device's camera and microphone.
Turn any idea into an investable index using Generated Assets. Enter a prompt and AI will pull a custom list of stocks to build an index tailored to your vision. Analyze performance, compare to the S&P 500, fine-tune holdings, and share it with the world. Create your own Generated Asset.*
Notion is significantly upgrading its AI toolkit. They've announced AI-powered meeting notes, enterprise search capabilities, a new research mode, a model picker to choose your preferred AI, and updated AI pricing. Notion vs Granola - im not affiliated with either but i’m not a notion maxi. so i’d pick granola every day.
related: i used notion mail for a while but no mobile app, a few funky bugs and im back to using superhuman.
*sponsored
want to partner with us? Click here
🌐 What I’m consuming
João Moura, founder of Crew AI (I'm an investor), on what really matters for AI agents. Spoiler: it’s production readiness and good engineering, not just frameworks.
This piece on AI knowing us too well asks if we really want AI to hold decades of our personal history.
Josh, founder of The Browser Company (makers of ARC) shared some lessons from building Dia, their new AI browser. Chat is great, memory is hard and “context” is the secret key. Interesting timing with Perplexity's Comet browser supposedly launching in 3-5 weeks.
DeepMind's AlphaEvolve designing advanced algorithms using Gemini, with new progress in open math problems, saving 0.7% of Google’s compute, making Gemini training 1% faster.
Google's AI Futures Fund is now live, already backing 12 AI startups.
Deep dive into building and scaling ChatGPT Images with the OpenAI team.
Which model should you choose in Cursor (and no, it’s not just ‘whatever one gets it to work’ - although, it is for a lot of users I’m sure. guilty)
⚙️ Tools I’m tinkering with
Exa Websets - Access massive datasets from the web, updated weekly with over 50k results. This is one of my favourite new tools to use. Some of the classification isn’t 100% yet, but still good.
Dots - New Computer's product, pitched as "the first social intelligence”. I know Sam, and met her and Jason when in SF. They’re very thoughtful about the experiences they build with AI. Excited to test a social network built with AI from the ground up. Early access coming soon…
Granola 2.0 - Now a "second brain for your team" with shared folders, chat with folders, and more, backed by a $43M Series B. If you’re using anything else, just drop it and get granola - trust me.
Lex desktop app - The AI writing tool I use now has a native desktop version. (im an investor)
Muscle Mem - Open-source cache to speed up AI agents by reducing LLM calls for repetitive tasks. (GitHub)
StageHand by BrowserBase can be used by CrewAI’s agents now.
Kithbook - A modern rolodex for busy founders, from a Ben's Bites community member.
Tensorlake Cloud - Managed platform to turn messy documents into clean, AI-ready structured data.
Minions Secure Chat - Encrypted LLM chats with models hosted on the cloud with low latency overhead.
Parseltongue - Translate text into 'unhinged' languages to test content filters by Pliny, the Elder god of jailbreaking LLMs.
📈 Why more companies should be running funds
A few weeks ago, Gumroad acquired Small Bets, a profitable, community-first business built by Daniel Vassallo.
Buried in Vassallo’s announcement was this:
“Small Bets will act as a laboratory… we’ll try to scale this idea of supporting builders.”
This stands out to me because it signals Gumroad’s strategic direction. Rather than treating builders just as users, they’re going to start actively helping them.
Sounds like a pretty altruistic move, but it’s deeply strategic.
More and more companies are recognising that their success is tied to the success of the businesses built on their tool.
Don’t trust me? I come with receipts in my full post →
🍦 Afters
YC's AI Startup School in SF - a free, in-person event for 2500 CS students
Demo Day video for Create’s Season 1 of building is live (port co of mine)
Reflect, a startup aiming to bring sunlight after dark with orbital lighting, raised a $20M Series A. How cool is that?!
AI roles are up ~2.5x in the last two years
Companies in Embed by Conviction's latest batch
That’s it for today. Feel free to hit reply and share your thoughts. 👋
Enjoy this newsletter? Please forward to a friend.