Once, I told a librarian that she'd be out of a job because all books would soon be on the internet. I was 8. San Francisco via Chicago via Costa Rica.
San Francisco, CA
Joined November 2008
- Tweets18.3K
- Following986
- Followers1.5K
- Likes34.3K
Pinned Tweet
Just posted my annual reading recap.
Traveling from SF to Mexico City, Vienna, Budapest, and South Africa, I circled one theme: ideas rewrite the world.
Standouts: Zweig on collapse, MacKenzie on financial models that shape markets.
Link to the blog below. What did you read?
Do I know any great consumer growth people? Do you know someone I should know?
I'm building a new AI fintech and looking for a founding hire in SF to get it into people's hands. Someone who's all-in on agents, likes the messy early days and figuring out what works. 💰
DMs open.
Whether your browser auto-completes g to gmail or github says a lot about who you are and how you spend your time at work.
Avy Faingezicht retweeted
You’re 25x more likely to be a millionaire than be homeless in the US
everyone thinks for some reason that they are closer to becoming a millionaire than becoming homeless.
Readers added context they thought people might want to know
Recent polling shows a majority of US voters believe they are more likely to become homeless than a millionaire. US data show roughly 23.6 million millionaires versus ~746,000 experiencing homelessness on a given night.
dataforprogress.org/blog/2026/8/13…
investopedia.com/new-report-rev…
hud.gov/news/hud-no-26…
Avy Faingezicht retweeted
Moats in 2026:
- having conviction for more than 5 days
- thinking for yourself
In the age of AI agents writing most of the code, do we care about test coverage, or is that also out the window?
Great things take time.
Our work, like capital, compounds. Congrats, @Erik_Syvertsen and the @usvc_ team!
For 15+ years I've worked with venture funds: 10-year fund lives and waiting for DPI. Today, @usvc_ 's Q3 Schedule TO arrived at my house offering to buy back my shares at NAV. Every shareholder got this same offer. Worth every late night.
A subtle difference I've noticed in Codex in the last couple of weeks is that they're hiding git branching and worktrees from the UI, letting the agents make choices about source control more and more.
This looks very cool. Pointers in your LLM tokens!
You already gave the LLM the answer. So why is it writing it out from scratch?
Introducing a new AI model architecture: Constrained Decoding Models (CDMs)
LLMs were built to write. Most production workloads just need an answer: fast, consistent, and affordable.
CDMs are purpose-built for 1 task each. They read your input once and skip the token-by-token generative loop that makes LLMs slow and expensive.
Not just a decision model, but a family of 3 CDMs live today:
• Classify: define your labels and get back a calibrated probability + reason for each. Used for routing, tagging, and intent detection.
• Extract: describe the entities in natural language and get back well-formed JSON with confidence scores + citations. Used for web scraping, document processing, and PII redaction.
• Summarize: pass in long text, images, or PDFs and get back a consistent summary under 1,000 tokens. Used for context compaction, metadata generation, and transcript digests.
Built for production:
• 1M tokens of multimodal (text + image) input, 31x larger than Jev
• 77.2% on DecisionBench vs Jev's 72%
• 200x faster, 467x cheaper than frontier LLM APIs
• $0.05/M input tokens. Output is free (yes, even for summarize)
Live on your favorite platforms today:
@OpenRouter @vercel_dev @LiteLLM @LangChain @TrustedRouter @itsArthurAI @useRouterPlus
Get 50M input tokens for free (for the first 50 hours) at link below 👀
There's a lot of hype with around what agents will inevitably end up doing for us. There's also a wide gap between what they _can_ do and what we trust them to do for us.
In that gap, there's opportunity.
We live in an age of wonders chatgpt.com/share/6acbaa15-b…
I can't imagine what it's like to be in college learning these things for the first time with these tools at hand.
I remember slaving over this Kuramoto simulation with @leonsasson back in ~2012, pulling and all nighter, and not really getting how it worked. But now you can just type two sentences and simulate this five different ways youtube.com/watch?v=jlCWNpIS…
The bots are trained to use Chrome, but I'm a Firefox guy, which gets both Codex and Claude confused about browsers, credentials, etc. RLHF runs the show.
Found this in my email, from 2015. A project scraping AngelList job postings to compare startup salary and equity ranges across cities. Some things never change.
The AI labs may end up taxing every application, but they won't own every business. I'm betting against the everything company.
New blog post, link below.
Avy Faingezicht retweeted
It is hard to articulate to people outside of tech how much working in tech has changed this year
It’s honestly kind of hard to believe
